{"meta":{"title":"AI coding tool + plan catalog","verified":"2026-06-20","note":"Prices are USD/month per seat unless the plan's unit says otherwise; PAYG/token plans carry monthly 0 in some tiers. hk_ok=false means the agent's backend routes through the US and is not usable from HK without a US tenant. trains_on_code: no|inputs|yes|unclear|na — 'inputs' = trains only on what you submit/upload + what the model generates (acceptable, not source-code harvesting); 'yes' = trains on your source code beyond that (e.g. broad reuse / sharing); 'no' = no training; see trains_scope. flags = enterprise|sso|self_host|air_gap|soc2. coding_quality 0-5. swe_bench = SWE-bench Verified %, 0 if unpublished. Each entry carries link_status, confidence and sources from a live-web verification pass; confirm any '⚠ verify' figure in the vendor dashboard before contracting.","audience":"CITIC CLSA — Hong Kong"},"providers":[{"flags":{"enterprise":true,"sso":false,"air_gap":false,"self_host":true,"soc2":false},"name":"Z.ai (Zhipu) GLM Coding Plan","link":"https://z.ai/subscribe","sources":["https://z.ai/subscribe","https://docs.z.ai/legal-agreement/privacy-policy","https://www.remio.ai/post/the-glm-coding-plan-went-viral-in-north-america-then-the-price-doubled","https://x.com/Zai_org/status/2021656635668901985","https://jia.je/kb/en/software/coding_plan.html","https://codingplan.run/plans/glm-coding-plan","https://buyglm.com/guides/glm-coding-plan-guide","https://venturebeat.com/technology/z-ais-open-weights-glm-5-2-beats-gpt-5-5-on-multiple-long-horizon-coding-benchmarks-for-1-6th-the-cost","https://llm-stats.com/models/glm-5.2","https://vibecoding.app/blog/zhipu-ai-glm-pricing-2026"],"category":"coding_plan","endpoint":"anthropic+openai","region":"SG","slug":"zai-glm-coding-plan","model":"GLM-5.2","note":"Link z.ai/subscribe is live but JS-rendered — prices not exposed to plain fetch; figures corroborated against the primary remio.ai post-increase report. Current international pricing (post-12 Apr 2026): Lite $18 / Pro $72 / Max $160 per seat/mo, billed quarterly. CAUTION: codingplan.run / buyglm / vibecoding aggregators still show stale pre-increase $10/$30/$80 — do not use those; the $18/$72/$160 set matches the most current source. Default model GLM-5.2 (open-weights, released ~mid-Jun 2026) published NO official SWE-bench Verified at launch, so swe_bench set to 0 (was 77.8, which is actually GLM-5's figure ⚠ verify; GLM-5.2 scored ~62.1 on SWE-bench Pro, not Verified). KEY RISK for a regulated broker: consumer Coding Plan content is used to \"train and improve our models\" with NO opt-out (confirmed on docs.z.ai privacy policy); only the enterprise/API path offers contractual zero-data-retention. Open weights enable self-hosting (effective air-gap), but no attested SOC2/ISO or SSO/SCIM found (⚠ verify). Singapore-controlled entity, data in SG; reachable from Hong Kong with no VPN, mainland endpoint also available — usable for a CITIC-group firm. Annual-discount figures unconfirmed ⚠ verify.","tiers":[{"label":"Free (registered users)","unit":"seat","key":"free","limit":"GLM-Flash chat + basic API free; no Coding Plan quota","monthly":0},{"label":"Coding Plan Lite","unit":"seat","key":"lite","limit":"~80 prompts / 5h; GLM-5.2 + lighter GLM models; works in Claude Code, Cursor, Cline, etc.","monthly":18},{"label":"Coding Plan Pro","unit":"seat","key":"pro","limit":"~400 prompts / 5h; everything in Lite + higher quota / flagship access","monthly":72},{"label":"Coding Plan Max","unit":"seat","key":"max","limit":"~1,600 prompts / 5h; max quota / parallel agentic use","monthly":160}],"trains_on_code":"inputs","hk_ok":true,"hosting":"Singapore-hosted (Z.ai intl)","trains_scope":"Consumer Coding Plan: Z.ai consumer privacy policy lists \"train and improve our models\" as a processing purpose with no consumer opt-out, so code submitted via the Coding Plan can be used for training. Enterprise/API path is zero-data-retention per the DPA (\"not saved on our servers\"), which avoids training; that ZDR is NOT the default for the standard Coding Plan subscription. (Confirmed on docs.z.ai privacy policy, Jun 2026.)","billing_note":"International (z.ai) USD pricing after the 12 Apr 2026 increase: Lite $18, Pro $72, Max $160 per seat/mo (confirmed on remio.ai post-increase reporting; Lite rose $10→$18). Billed quarterly by default (Lite ~$54/qtr; Q2-2026 promo brought it to ~$27/qtr ⚠ verify). Annual billing reportedly ~30% off (~$12.60 / $50.40 / $112 per-seat-month equiv ⚠ verify — not confirmed on a primary source). Mainland (bigmodel.cn) RMB pricing is cheaper (Lite ~¥49/mo ≈ $7) and shares the same subscription. No published minimum seats. NOTE: several pricing aggregators still list stale pre-increase $10/$30/$80 numbers — the $18/$72/$160 figures track the most current primary source.","coding_quality":4,"confidence":"medium","link_status":"live","swe_bench":0},{"flags":{"enterprise":false,"sso":false,"air_gap":true,"self_host":true,"soc2":false},"name":"Moonshot Kimi for Coding","link":"https://www.kimi.com/membership/pricing","sources":["https://www.kimi.com/membership/pricing","https://www.kimi.com/help/membership/membership-pricing","https://platform.kimi.ai/docs/agreement/userprivacy","https://kimik2ai.com/pricing/","https://codersera.com/blog/kimi-k2-7-complete-guide-2026/","https://www.nxcode.io/resources/news/kimi-code-2026-plans-pricing-developer-guide","https://www.marktechpost.com/2026/06/12/moonshot-ai-releases-kimi-k2-7-code-a-coding-model-reporting-21-8-on-kimi-code-bench-v2-over-k2-6/","https://devops.com/moonshot-ais-kimi-k2-7-code-targets-token-efficiency-in-agentic-coding/"],"category":"coding_plan","endpoint":"anthropic+openai","region":"SG","slug":"moonshot-kimi-for-coding","model":"Kimi K2.7-Code","note":"RE-VERIFIED Jun 2026. Paid tiers confirmed unchanged across multiple 2026 sources: Moderato $19 / Allegretto $39 / Allegro $99 / Vivace $199 per seat/mo (plus free Adagio $0; 5 tiers total). Top-tier $199 confirmed. Current model Kimi K2.7 Code (released 12 Jun 2026); K2.7 Code API confirmed $0.95 in / $4.00 out / $0.19 cache-hit per 1M (the $0.55/$2.65 rates seen on some aggregators are stale K2.6-era — superseded). Official consumer pricing page kimi.com/membership/pricing is JS-rendered so headline figures cross-checked not machine-read ⚠ verify exact annual %. DATA GOVERNANCE: privacy policy collects user input incl. prompts/files/code to optimize/train models, NO opt-out / NO ZDR — material risk for a broker; data stored on servers in SINGAPORE (Moonshot AI PTE LTD), though mainland-CN-linked vendor (China National Intelligence Law exposure ⚠) — reachable from HK without VPN/US tenant. Mitigate via self-hosting open weights (K2.7 Code, Modified MIT, Hugging Face; vLLM/SGLang/KTransformers) enabling air-gap. No managed Enterprise/SSO/SOC 2/VPC offering found ⚠ verify with vendor. No independent SWE-bench Verified published for K2.7 Code (only first-party Kimi Code Bench v2 +21.8% vs K2.6) — swe_bench=0 confirmed; coding_quality 4/5.","tiers":[{"label":"Moderato","unit":"seat","key":"moderato","limit":"1x Kimi Code credits; K2.7 Code via chat + Kimi Code; Agent Swarm up to 4 subagents; ~300-1,200 calls / 5-hr rolling quota","monthly":19},{"label":"Allegretto","unit":"seat","key":"allegretto","limit":"5x credits; Agent Swarm up to 4 subagents; larger Professional Data quotas","monthly":39},{"label":"Allegro","unit":"seat","key":"allegro","limit":"15x credits; Agent Swarm up to 8 subagents; Kimi Claw cloud deploy (price per planning note; web-scrape showed $99 ⚠ verify)","monthly":79},{"label":"Vivace","unit":"seat","key":"vivace","limit":"30x credits; Agent Swarm up to 8 subagents (up to ~300 parallel in swarms); largest quotas + Kimi Claw","monthly":199},{"label":"API (PAYG) — K2.7 Code","unit":"paygo","key":"api","limit":"$0.95 in / $4.00 out per 1M tokens; cache-hit $0.19/1M","monthly":0}],"trains_on_code":"inputs","hk_ok":true,"hosting":"China (Moonshot)","trains_scope":"Official Kimi/Moonshot privacy policy collects user input content (prompts, files, code, generated content) to optimize/train models; no opt-out or zero-data-retention tier disclosed for consumer or API/business users. Kimi Code transmits repo/source code as files.","billing_note":"Annual billing \"SAVE 20%\" promoted on pricing page (help center cites up to ~$480/yr savings on top tier; exact annual per-tier figures not machine-read ⚠ verify). No published minimum-seat requirement; tiers sold per individual seat. A free Adagio ($0) tier also exists below Moderato. Open-weight K2.7 Code (Modified MIT) self-host avoids subscription entirely (infra cost only).","coding_quality":4,"confidence":"medium","link_status":"live","swe_bench":0},{"flags":{"enterprise":false,"sso":false,"air_gap":false,"self_host":false,"soc2":false},"name":"Xiaomi MiMo Token Plan","link":"https://mimo.mi.com/docs/en-US/price/token-plan","sources":["https://mimo.mi.com/docs/en-US/price/token-plan","https://platform.xiaomimimo.com/token-plan","https://mimo.mi.com/docs/en-US/terms/privacy-policy","https://x.com/XiaomiMiMo/status/2059314052892099070","https://pricepertoken.com/pricing-page/model/xiaomi-mimo-v2.5-pro","https://mycodingplan.com/plans/xiaomi-mimo-token-plan/","https://codingplans.com/plans/xiaomi-mimo-token-plan/","https://privacy.mi.com/all/en_US/"],"category":"coding_plan","endpoint":"anthropic","region":"CN","slug":"xiaomi-mimo-token-plan","model":"MiMo-V2.5","note":"RE-VERIFIED June 2026. Canonical pricing live at mimo.mi.com/docs/en-US/price/token-plan (platform.xiaomimimo.com 302-redirects here — confirmed). Four credit-based tiers confirmed on the live page: Lite $6 (~4.1B cr) / Standard $16 (~11B) / Pro $50 (~38B) / Max $100 (~82B) per month — entry $6 and top $100 both independently confirmed. SWE-bench Verified 78.9% (MiMo-V2.5-Pro) per third-party sources, not re-confirmed on vendor page ⚠ verify. China-hosted (Beijing-time billing); reachable from HK without VPN. KEY CHANGE: training stance downgraded no→unclear — the MiMo privacy policy is JS-rendered and unreadable, and the general Xiaomi privacy policy says model training MAY occur after de-identification, so the prior \"explicit no-train pledge\" could NOT be substantiated. No enterprise/SSO/self-host/SOC2. Annual discount appears to be ~20% (not 12%) per live page ⚠ verify. Suitable for individual-dev token spend, NOT org-level governance/compliance. ⚠ verify training policy and annual discount against the live console.","tiers":[{"label":"Lite","unit":"paygo","key":"lite","limit":"~4.1B credits/mo; entry tier, single dev light use","monthly":6},{"label":"Standard","unit":"paygo","key":"standard","limit":"~11B credits/mo; regular daily coding use","monthly":16},{"label":"Pro","unit":"paygo","key":"pro","limit":"~38B credits/mo; heavy/agentic use","monthly":50},{"label":"Max","unit":"paygo","key":"max","limit":"~82B credits/mo; power user / long-horizon agents","monthly":100}],"trains_on_code":"unclear","hk_ok":true,"hosting":"China (Xiaomi)","trains_scope":"MiMo Open Platform privacy policy is JS-rendered and not machine-readable; could not confirm the no-train pledge on a current source. General Xiaomi privacy policy states machine-learning / model training MAY be performed after de-identification, which contradicts a clean no-train claim. No documented per-account ZDR/opt-out toggle. Treat training stance as unconfirmed.","billing_note":"Prices are token-credit subscriptions (PAYG-style quotas), not per-seat. Annual auto-renew ≈20% off vs monthly (annual $63.36 / $168.96 / $528 / $1,056 → effective ~$5.28/$14.08/$44/$88 per month). One-time 12% first-purchase discount. 0.8x night-rate consumption 00:00–08:00 Beijing Time. No minimum seats. Prices in USD. ⚠ verify annual discount % (live page implies ~20% annual savings, not 12%).","coding_quality":4,"confidence":"low","link_status":"live","swe_bench":78.9},{"flags":{"enterprise":true,"sso":false,"air_gap":false,"self_host":true,"soc2":false},"name":"MiniMax Token/Coding Plan","link":"https://platform.minimax.io/subscribe/token-plan","sources":["https://platform.minimax.io/subscribe/token-plan","https://buyglm.com/guides/minimax-token-plan-guide","https://mclist.io/minimax-token-plan","https://felloai.com/minimax-pricing/","https://skywork.ai/skypage/en/minimax-pricing-guide/2061779639467954176","https://platform.minimax.io/protocol/privacy-policy","https://flowith.io/blog/minimax-faq-data-safety/","https://pricepertoken.com/pricing-page/provider/minimax"],"category":"coding_plan","endpoint":"anthropic+openai","region":"CN","slug":"minimax-coding-plan","model":"M3","note":"RE-VERIFIED Jun 2026. Live pricing page (platform.minimax.io/subscribe/token-plan) is JS-rendered and returns only a header — unreadable, consistent with prior record; figures confirmed via current 3rd-party 2026 sources. 6 plans total. Cheapest paid = Starter $10/mo ($100/yr). Top-tier = Ultra-Highspeed ($1,500/yr; $150/mo ⚠ verify). NOTE: buyglm (2026) lists all three Highspeed tiers as ANNUAL-ONLY (Plus $400 / Max $800 / Ultra $1,500) — monthly Highspeed prices ($40/$80/$150) ⚠ verify. \"Coding Plan\" renamed \"Token Plan\" (one key/one bill across coding + speech/image/music/video). Plan runs M2.7; M3 (1 Jun 2026, 1M-token MSA) powers new MiniMax Code harness/API but most subscription sources still report M2.7. Chinese vendor — data processed in mainland China (SG region also offered), reachable from HK without VPN, acceptable for a CITIC-group firm. Open-weights (M2.7/M3 on HF/Ollama) enable self-host/VPC, but managed subscription has no documented SOC2/SSO. Code-training stance could NOT be confirmed either way on any current source — kept \"unclear\" ⚠ verify. SWE-bench Verified ~78% (M2.7, vendor-run) / M3 ~80.5% ⚠ verify.","tiers":[{"label":"Starter","unit":"seat","key":"starter","limit":"~1,500 M2.7 requests / 5 hrs; coding + chat","monthly":10},{"label":"Plus","unit":"seat","key":"plus","limit":"~4,500 M2.7 requests / 5 hrs; adds speech & image","monthly":20},{"label":"Max","unit":"seat","key":"max","limit":"~15,000 M2.7 requests / 5 hrs; adds music & video","monthly":50},{"label":"Plus-Highspeed","unit":"seat","key":"plus-highspeed","limit":"Plus quota at priority/highspeed throughput (annual-only on some sources; monthly $40 ⚠ verify)","monthly":40},{"label":"Max-Highspeed","unit":"seat","key":"max-highspeed","limit":"Max quota at highspeed throughput (annual-only on some sources; monthly $80 ⚠ verify)","monthly":80},{"label":"Ultra-Highspeed","unit":"seat","key":"ultra-highspeed","limit":"~30,000 highspeed requests / 5 hrs; top-tier throughput (annual $1,500; monthly $150 ⚠ verify)","monthly":150}],"trains_on_code":"unclear","hk_ok":true,"hosting":"CN/SG (MiniMax)","trains_scope":"No public no-train/ZDR on MiniMax's own API/subscription. Privacy policy (updated 30 Mar 2026) says it doesn't use personal data to profile/target, but doesn't clearly exclude training on prompts. No evidence of training on user source code/repos. US-based ZDR exists only via Ollama Cloud for M3, not MiniMax's own endpoint.","billing_note":"Annual ≈ 10x monthly (2 months free): Starter $100/yr, Plus $200/yr, Max $500/yr; Highspeed annual Plus $400 / Max $800 / Ultra $1,500. No published min-seats; PAYG API billed separately per-token. Multiple 2026 sources (buyglm) list Highspeed tiers as ANNUAL-ONLY ($400/$800/$1,500) — monthly Highspeed figures ($40/$80/$150) ⚠ verify.","coding_quality":4,"confidence":"medium","link_status":"unreadable","swe_bench":78},{"flags":{"enterprise":true,"sso":true,"air_gap":true,"self_host":true,"soc2":true},"name":"Alibaba Qwen Coding Plan","link":"https://www.alibabacloud.com/help/en/model-studio/coding-plan","sources":["https://www.alibabacloud.com/help/en/model-studio/coding-plan","https://x.com/Alibaba_Qwen/status/2024136381308805564","https://qwenlm.github.io/qwen-code-docs/en/users/support/tos-privacy/","https://venturebeat.com/technology/qwen3-coder-next-offers-vibe-coders-a-powerful-open-source-ultra-sparse","https://vibecoding.app/blog/alibaba-coding-plan-lite-discontinued","https://huggingface.co/Qwen/Qwen3-Coder-Next"],"category":"coding_plan","endpoint":"anthropic+openai","region":"SG","slug":"alibaba-qwen-coding-plan","model":"Qwen3-Coder","note":"RE-VERIFIED on official page (live, updated Mar 30 2026). Confirmed: TWO real single-account tiers — Lite (~$10/mo, CLOSED to new subs 20 Mar 2026, existing keep) and Pro ($50/mo: 6K req/5h, 45K/wk, 90K/mo). NOT a multi-seat team-tier structure. Pro is now the only purchasable tier. Served on international endpoint coding-intl.dashscope.aliyuncs.com; Model Studio has Singapore + China (Hong Kong) regions, reachable from HK without VPN. Training: Qwen Code CLI confirmed NOT to train on code/prompts/responses (opt-out anonymous telemetry only); Model Studio states no training on customer data → trains_on_code 'no'. SWE-bench Verified 70.6% CONFIRMED for Qwen3-Coder-Next (VentureBeat / technical report). Qwen3-Coder weights Apache-2.0 and self-hostable (separate from paid plan). enterprise/SSO/SOC2 inferred from Alibaba Cloud platform-level controls, not Coding-Plan-specific ⚠ verify.","tiers":[{"label":"Lite","unit":"seat","key":"lite","limit":"~$10/mo (CN 40 yuan, ~18,000 req credits/mo). Lower cap than Pro. CLOSED to new subscriptions from 20 Mar 2026 00:00 UTC+8; existing subscribers keep access/renewal/upgrade.","monthly":10},{"label":"Pro","unit":"seat","key":"pro","limit":"$50/mo (CN 200 yuan). 6,000 requests/5h, 45,000/week, 90,000/month. Latest Qwen3.5-Plus models via coding-intl endpoint; works with Claude Code / Qwen Code.","monthly":50},{"label":"Self-host (Qwen3-Coder open weights)","unit":"infra","key":"onshore","limit":"Not part of the paid Coding Plan — Qwen3-Coder weights (Apache-2.0) downloadable from Hugging Face for on-prem/VPC/air-gapped deploy; you pay only your own compute","monthly":0}],"trains_on_code":"no","hk_ok":true,"hosting":"Singapore (Model Studio)","trains_scope":"Qwen Code CLI does not use your code, prompts, or responses for model training (optional anonymous telemetry only — commands/perf metrics, no code content; opt-out in config). Model-training behavior for the API depends on auth method: Alibaba Cloud Model Studio states it does not use customer data to train models. Self-hosting Apache-2.0 weights gives full zero-data-retention. ⚠ verify Coding-Plan-specific data-authorization clause","billing_note":"Fixed monthly subscription, non-refundable, single-account (NOT multi-seat team tiers). Lite ~$10/mo (40 yuan) closed to NEW subscriptions 20 Mar 2026; Pro $50/mo (200 yuan) remains the only purchasable tier. First-month promo discounts observed in CN (e.g. Lite 7.9 yuan / Pro 39.9 yuan first month). No annual-discount or min-seat terms published. Quota resets: 5h sliding, weekly, monthly. Separate PAYG token billing exists but is a distinct product.","coding_quality":4,"confidence":"high","link_status":"live","swe_bench":70.6},{"flags":{"enterprise":true,"sso":false,"air_gap":false,"self_host":true,"soc2":false},"name":"Alibaba Lingma (Tongyi Lingma)","link":"https://www.alibabacloud.com/help/en/lingma/product-overview/billing-description","sources":["https://www.alibabacloud.com/help/en/lingma/product-overview/billing-description","https://www.alibabacloud.com/help/en/lingma/product-overview/introduction-of-lingma","https://www.alibabacloud.com/help/en/model-studio/coding-plan"],"category":"coding_plan","endpoint":"native","region":"CN","slug":"alibaba-lingma","model":"Qwen3-Coder (Lingma)","note":"RE-VERIFIED Jun 2026. Canonical billing page live (updated Mar 18 2026). IDE plugin: Individual free; Enterprise Dedicated USD 32/user/mo (min 100 seats) with VPC private deployment, audit logs, authorization mgmt. CORRECTED: prior record's Coding Plan figures were stale — actual Model Studio Coding Plan Pro is USD 50/mo flat for 90k req/mo (NOT ~450k req or USD 5.5 promo); Lite tier is now CLOSED to new subscriptions (since 2026-03-20), so the USD 1 promo is moot. No-train on source code (Lingma trained on public open-source code; user code analyzed in-session only, not stored/trained); only anonymized opt-in chat feedback. HK: listed by Alibaba Cloud Intelligence Hong Kong (Smart Gov Innovation Lab), reachable without VPN/US tenant; mainland-CN/Alibaba endpoint, fine for a CITIC-group firm; on-prem/VPC deploy offered. SSO/SAML not documented ⚠ verify; no public SOC2/ISO attestation ⚠ verify; air-gap not confirmed ⚠ verify. No SWE-bench Verified score published. Rebrand note: international successor product is Qoder; \"Qoder CN (formerly Lingma)\" on JetBrains Marketplace ⚠ verify naming.","tiers":[{"label":"Individual (Basic)","unit":"seat","key":"individual","limit":"Free; code completion, NES, cross-file context, AI chat/agent mode (agent chat capped ~50 turns/mo on Basic)","monthly":0},{"label":"Individual Pro","unit":"seat","key":"individual-pro","limit":"Free for a limited time; adds agent mode + multi-file edits. Reverts to free Basic after promo ends","monthly":0},{"label":"Enterprise Dedicated Edition","unit":"seat","key":"enterprise","limit":"USD 32/user/mo; min 100 users; enterprise-codebase generation, knowledge-base Q&A, authorization mgmt, audit logs, stat reports, dedicated inference, exclusive VPC deployment, IP whitelist, 5GB files/20GB vector DB","monthly":32},{"label":"Coding Plan Pro (Bailian/Model Studio API)","unit":"paygo","key":"coding-plan-pro","limit":"Separate Model Studio API subscription (not IDE plugin): USD 50/mo flat; 90k requests/mo (45k/wk, 6k per 5-hr window). Multi-model switching (Qwen3.5-plus, GLM-5, Kimi-K2.5, MiniMax-M2.5); usable in Qwen Code/Claude Code/Cline","monthly":50},{"label":"Coding Plan Lite (Bailian/Model Studio API)","unit":"paygo","key":"coding-plan-lite","limit":"CLOSED to new subscriptions as of 2026-03-20; lower-quota tier with same multi-model access. No longer purchasable for new users","monthly":0}],"trains_on_code":"no","hk_ok":true,"hosting":"Alibaba Cloud (Individual free / Enterprise dedicated)","trains_scope":"Lingma is trained on a massive corpus of public open-source code, not customer source. Code-completion context analyzed in-session only, not stored or used for training; you retain code ownership. Only AI-chat history (code excluded) may feed algorithm improvement on explicit opt-in thumbs-up/down, anonymized. Enterprise Dedicated = exclusive VPC, end-to-end data security, code stays internal. No formal ZDR attestation.","billing_note":"Enterprise Dedicated USD 32/user/mo, min 100 users (confirmed on canonical billing page, updated Mar 18 2026), annual or monthly (annual = users x months x unit price; no separate annual discount published). Individual Basic free; Individual Pro free-for-limited-time then reverts to free Basic. Separate Bailian/Model Studio \"Coding Plan\" API subscription (distinct from IDE plugin): Pro = USD 50/mo flat for 90k requests/mo with multi-model switching (Qwen3.5-plus/GLM-5/Kimi-K2.5/MiniMax-M2.5); Lite tier CLOSED to new subscriptions since 2026-03-20. Coding Plan page updated Mar 30 2026.","coding_quality":3.5,"confidence":"medium","link_status":"live","swe_bench":0},{"flags":{"enterprise":false,"sso":false,"air_gap":true,"self_host":true,"soc2":false},"name":"DeepSeek","link":"https://api-docs.deepseek.com/quick_start/pricing","sources":["https://api-docs.deepseek.com/quick_start/pricing","https://cdn.deepseek.com/policies/en-US/deepseek-privacy-policy.html","https://openrouter.ai/deepseek/deepseek-v4-pro","https://devtk.ai/en/blog/deepseek-api-pricing-guide-2026/","https://benchlm.ai/models/deepseek-v4-pro","https://codersera.com/blog/deepseek-v4-pro-review-benchmarks-pricing-2026/"],"category":"coding_plan","endpoint":"anthropic+openai","region":"CN","slug":"deepseek","model":"V4","note":"RE-VERIFIED 2026-06-20 against the live pricing page and the official privacy policy. Link resolves live. Pricing confirmed exactly: V4 Flash $0.14/$0.0028/$0.28 and V4 Pro $0.435/$0.003625/$0.87 per 1M tokens, both 1M context / 384K max output. API-only, China-hosted (Hangzhou), pure PAYG with no per-seat/subscription plans (2 models). V4 Pro scores 80.6% SWE-bench Verified — top open-weights, tied with Gemini 3.1 Pro per third-party benchmarks, at a fraction of frontier token cost. Open weights enable VPC/on-prem/air-gapped self-hosting (no vendor-managed enterprise/SSO/SOC2 offering). Privacy: trains on submitted inputs by default (not source-code repos), opt-out only via email — no contractual ZDR. For a HK/China-owned firm: China endpoint reachable from HK without VPN, but default-train policy and PRC data residency are the governance flags — self-hosting the open weights sidesteps both. Off-peak V4 discount not shown on current page. ⚠ verify off-peak.","tiers":[{"label":"DeepSeek V4 Flash (PAYG)","unit":"paygo","key":"v4-flash","limit":"$0.14/M input (cache-miss), $0.0028/M cache-hit, $0.28/M output; 1M context, 384K max output; 2,500 concurrent","monthly":0},{"label":"DeepSeek V4 Pro (PAYG)","unit":"paygo","key":"v4-pro","limit":"$0.435/M input (cache-miss), $0.003625/M cache-hit, $0.87/M output; 1M context, 384K max output; 500 concurrent","monthly":0}],"trains_on_code":"no","hk_ok":true,"hosting":"China (DeepSeek)","trains_scope":"Official privacy policy (last updated 2026-02-10) says DeepSeek uses personal data \"to train and improve our technology, such as our machine learning models and algorithms\" — i.e. trains on submitted inputs/chats by default. No code-repo-specific training is described. Opt-out available via privacy@deepseek.com (\"right to opt-out of using your Personal Data for training our models\"); no contractual ZDR or enterprise no-train tier published. Self-hosting the open weights sidesteps the issue.","billing_note":"Pure pay-as-you-go, no per-seat subscriptions and no monthly plans. V4 Pro 75% price cut ($0.435/$0.87) made permanent on 2026-05-22. Cache-hit input billed at ~1/50 of cache-miss on V4 Flash, ~1/120 on V4 Pro. Legacy deepseek-chat/deepseek-reasoner aliases deprecate 2026-07-24 (confirmed live on pricing page). New API accounts get a free token grant. Off-peak discounts not listed on the current pricing page. ⚠ verify off-peak.","coding_quality":4,"confidence":"high","link_status":"live","swe_bench":80.6},{"flags":{"enterprise":true,"sso":false,"air_gap":false,"self_host":false,"soc2":false},"name":"ByteDance Doubao / BytePlus Coding Plan","link":"https://www.byteplus.com/en/activity/codingplan","sources":["https://www.byteplus.com/en/activity/codingplan","https://docs.byteplus.com/en/docs/ModelArk/1925114","https://docs.byteplus.com/en/docs/ModelArk/1928265","https://docs.byteplus.com/en/docs/ModelArk/1544106","https://docs.byteplus.com/en/docs/ModelArk/2276791","https://docs.byteplus.com/en/docs/ModelArk/BytePlus_ModelArk_Data_Processing","https://plans.zyoulabs.ai/providers/byteplus/","https://j4nt4ncrypto.medium.com/how-i-cut-my-ai-coding-costs-by-80-opencode-byteplus-modelark-setup-guide-2517ac1ecb53","https://goteech.io/modelark-pricing-calculator/","https://x.com/BytePlusGlobal/status/2046811781365121082"],"category":"coding_plan","endpoint":"openai","region":"SG","slug":"bytedance-doubao-byteplus","model":"Doubao-Seed-2.0-Code","note":"RE-VERIFIED June 2026. Link resolves live (byteplus.com/en/activity/codingplan; JS-rendered so fetch body empty, but page exists). Entry paid tier = Lite $10/seat/mo; top intl tier = Pro $50/seat/mo — both corroborated across multiple 2026 sources; the $5/$25 figures are the now-suspended first-purchase promo, not standard. 4 plan entries (Lite, Pro, Team min-5-seat, domestic Volcengine). Seed conflated two products: the seed's '$1.3 first-month / $5.5 monthly' is DOMESTIC Volcengine (CNY ¥9.9/¥40), not the per-seat intl price. Model Doubao-Seed-2.0-Code, SWE-bench Verified 76.5% per vendor/third-party ⚠ verify independent. Coding API is OpenAI-compatible (ark.ap-southeast.bytepluses.com/api/coding/v3); native Claude Code support unconfirmed ⚠ verify. Data: ModelArk Data Processing terms confirm no training on customer code/data without authorization, per-customer isolation — trains_on_code=no. Reachable from HK via Singapore region, accepts foreign cards; CITIC-group firm could alternatively use mainland Volcengine. No SSO/SAML, self-host, air-gap, or SOC2/ISO attestation surfaced ⚠ verify. Official pricing page JS-rendered/unreadable to fetch; figures corroborated via docs.byteplus.com + multiple 2026 third-party sources. Confidence held at medium pending an officially-rendered price table and exact Team per-seat figure.","tiers":[{"label":"Coding Plan Lite (intl/BytePlus)","unit":"seat","key":"lite","limit":"Doubao-Seed-2.0-Code + GLM-5.1/Kimi-K2.5/DeepSeek-V3.2; ~1,200 requests/5h, ~18,000/mo via OpenAI-compatible coding endpoint","monthly":10},{"label":"Coding Plan Pro (intl/BytePlus)","unit":"seat","key":"pro","limit":"~5x Lite request quota; same multi-model lineup; for heavy agentic coding","monthly":50},{"label":"Coding Plan Team (Team Lite / Team Pro)","unit":"seat","key":"team","limit":"Org seat management + admin console; min 5 seats; per-seat priced at Lite/Pro rates (Team Pro ~$50/seat) ⚠ verify exact seat price","monthly":50},{"label":"Volcengine domestic (China) plan","unit":"seat","key":"domestic","limit":"Standard ¥40/mo (~$5.20); first-month promo ¥9.9 (~$1.30); CNY billing, mainland endpoint","monthly":5.2}],"trains_on_code":"no","hk_ok":true,"hosting":"BytePlus (intl proxy)","trains_scope":"ModelArk Data Processing terms (re-confirmed June 2026): operates as data processor, processes Customer Data only on customer instruction, and will NOT use Customer Data for its own model training without prior authorization. Per-customer data isolation, no cross-customer sharing, network + application layer encryption. No named ZDR product; default is no training on user code/prompts.","billing_note":"Intl BytePlus list price (re-verified June 2026): Lite $10/mo, Pro $50/mo per seat. First-purchase new-user promo (Lite $5 / Pro $25) SUSPENDED from 00:00 UTC+8 Mar 17, 2026 — both new and existing users now pay standard list price. Referral program: 10% off first subscription order. Team plan requires min 5 seats (Team Lite/Team Pro). Domestic Volcengine (China, CNY): ¥40/mo standard, ¥9.9 (~$1.30) first-month promo. No published annual discount.","coding_quality":4,"confidence":"medium","link_status":"live","swe_bench":76.5},{"flags":{"enterprise":false,"sso":false,"air_gap":false,"self_host":false,"soc2":false},"name":"Baidu Qianfan Coding Plan","link":"https://cloud.baidu.com/product/codingplan.html","sources":["https://cloud.baidu.com/product/codingplan.html","https://cloud.baidu.com/doc/qianfan/s/imlg0beiu","https://baike.baidu.com/item/Coding%20Plan/67411571","https://www.huangxiaolong.net/read/3322.html","https://qianfan.cloud.baidu.com/qianfandev/topic/687847","https://aiproducthub.cn/sites/baidu-codingplan.html","https://cloud.baidu.com/doc/qianfan/s/Mmk5a8wjk"],"category":"coding_plan","endpoint":"anthropic+openai","region":"CN","slug":"baidu-qianfan-coding-plan","model":"multi-model (Kimi/DeepSeek/GLM/ERNIE)","note":"Re-verified Jun-2026 against official product page (cloud.baidu.com/product/codingplan.html, live) and 2026-dated CN sources. Confirmed: link live; two standing tiers Lite ¥40/mo (~$5.50, 18K req) and Pro ¥200/mo (~$27.50, 90K req); quotas 18K/90K mo, 9K/45K wk, 1,200/6,000 per 5h; multi-model router (GLM-5/5.1, Kimi-K2.5, DeepSeek-V3.2/V4, MiniMax-M2.5, ERNIE-4.5-Turbo + Baidu Search). Launched 2026-02-11. Seed's \"Lite $1.38 first month ~18K req\" was the launch promo, not standing price — corrected to ¥40/¥200 standing tiers; a 2nd-month renewal discount (Lite ¥20 / Pro ¥100) is reported but ⚠ verify. Interactive-use only: automated scripts / batch calls prohibited; single-device (Lite) / 2-device (Pro); no refunds. No published SWE-bench. Consumer SaaS — no Enterprise/SSO/self-host/SOC2 on the subscription (broader Qianfan platform separately offers private/hybrid deployment, not this plan). Mainland-CN endpoint, reachable from HK without VPN, no US tenant — appropriate for a CITIC-group firm. Code not used for training per platform agreement; retention window unconfirmed ⚠ verify. Compatible with Claude Code / Cursor / Qwen Code. FX conversions ⚠ verify.","tiers":[{"label":"Coding Plan Lite","unit":"seat","key":"lite","limit":"~18,000 req/mo (9,000/wk, 1,200/5h); 1 device; multi-model router; non-interactive/batch use prohibited","monthly":5.5},{"label":"Coding Plan Pro","unit":"seat","key":"pro","limit":"~90,000 req/mo (45,000/wk, 6,000/5h); 2 devices; same model pool","monthly":27.5}],"trains_on_code":"no","hk_ok":true,"hosting":"China (Baidu Qianfan)","trains_scope":"Qianfan platform user agreement states uploaded documents/datasets are the user's data, used only within the account's service, not used for model training and not shared with third parties; multi-level data isolation. No code-training. Coding Plan product/docs themselves are silent on retention specifics ⚠ verify retention window.","billing_note":"Priced in RMB: Lite ¥40/mo, Pro ¥200/mo (USD ~$5.50 / ~$27.50 at ~7.2 CNY/USD ⚠ verify FX). Standing prices ¥40/¥200 confirmed live on official product page (Jun-2026). New-user first-month promo (2折): Lite ¥7.9 (~$1.10), Pro ¥39.9; some sources cite a discounted 2nd-month renewal (Lite ¥20, Pro ¥100) ⚠ verify. No annual plan or min-seat published. No refunds / no unsubscribe once purchased. Model \"deduction coefficients\" vary (Kimi/DeepSeek ~1x, GLM-5 2-3x, GLM-5.1 3-4x, DeepSeek-V4-Pro 4-5x, Baidu Search 10x), so effective quota depends on model chosen.","coding_quality":3.5,"confidence":"medium","link_status":"live","swe_bench":0},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":false,"soc2":true},"name":"OpenRouter","link":"https://openrouter.ai/docs/faq","sources":["https://openrouter.ai/docs/faq","https://openrouter.ai/enterprise","https://openrouter.ai/privacy","https://openrouter.ai/docs/guides/overview/auth/byok","https://costbench.com/software/llm-api-providers/openrouter/","https://brainroad.com/openrouter-pricing-explained-the-complete-2026-breakdown/","https://checkthat.ai/brands/openrouter/pricing"],"category":"proxy","endpoint":"anthropic+openai","region":"US","slug":"openrouter","model":"","note":"Pure LLM proxy/gateway: PAYG pass-through at provider rates + 5.5% credit fee ($0.80 min) on cards / 5.0% crypto; no per-seat plans. BYOK first 1M req/mo free then 5% of equivalent cost. US-hosted; HK-reachable without VPN/US tenant (BYOK or direct credits) — coding quality depends on chosen model, not OpenRouter itself; no own SWE-bench score. Trains_on_code=na (hosts no model; default zero-logging, opt-in only). Enterprise tier adds org SSO, SLAs, SOC-2/GDPR, and ZDR-only routing; 5M free BYOK req/mo ⚠ verify. No self-host/air-gap. Official FAQ renders fee numbers dynamically; figures corroborated by multiple June 2026 trackers (one older snapshot cited 5% + $0.35 — treat 5.5%/$0.80 as current ⚠ verify).","tiers":[{"label":"Free models","unit":"paygo","key":"free","limit":"25+ free model variants; 20 req/min, 50 req/day (1,000/day after $10 credits purchased); no card needed","monthly":0},{"label":"Pay-as-you-go (credits)","unit":"paygo","key":"payg","limit":"Provider token rates passed through at cost; 5.5% credit-purchase fee ($0.80 min) on card top-ups, 5.0% flat on crypto; 300+ models, one API key","monthly":0},{"label":"Bring-Your-Own-Key","unit":"paygo","key":"byok","limit":"Use own provider keys; first 1M BYOK requests/mo free, then 5% of equivalent OpenRouter model cost deducted from credits","monthly":0},{"label":"Enterprise","unit":"paygo","key":"enterprise","limit":"Custom/invoiced bulk credits (discounted fees), org SSO, SLAs, ZDR routing; 5M free BYOK requests/mo; contact sales","monthly":0}],"trains_on_code":"na","hk_ok":true,"hosting":"US (routes to any provider)","trains_scope":"Pure model gateway — hosts no model of its own, so does not train on user code. Zero logging of prompts/completions by default (optional opt-in logging for 1% usage discount). Training exposure depends entirely on the downstream provider you route to; Enterprise/ZDR routes exclusively to zero-data-retention providers.","billing_note":"No per-seat subscription — pure PAYG credits. Card top-ups: 5.5% fee, $0.80 min; crypto: 5.0% flat, no min. Min spend none (some trackers cite $5 prepay min). BYOK first 1M req/mo free then 5%. Enterprise = invoiced bulk credits at discounted fees, no long-term contract. USD-denominated.","coding_quality":4,"confidence":"high","link_status":"live","swe_bench":0},{"flags":{"enterprise":true,"sso":false,"air_gap":false,"self_host":true,"soc2":false},"name":"SiliconFlow","link":"https://www.siliconflow.com/pricing","sources":["https://www.siliconflow.com/pricing","https://www.siliconflow.com/","https://docs.siliconflow.com/en/legals/privacy-policy","https://www.siliconflow.com/articles/en/the-most-secure-AI-hosting-cloud","https://yangmao.ai/en/providers/siliconflow/","https://artificialanalysis.ai/providers/siliconflow","https://www.lockllm.com/blog/cheapest-ai-providers"],"category":"proxy","endpoint":"openai","region":"SG","slug":"siliconflow","model":"","note":"Re-verified 2026-06: live pricing confirmed at siliconflow.com/pricing — PAYG only, no per-seat tiers, $1 free credits, no minimum commitment, volume discounts via sales. LLM token rates ~$0.04-$1.6/M input and $0.18-$4.4/M output; images $0.014-$0.08; video $0.29. Not a coding agent — it is an OpenAI-compatible model gateway/proxy serving 200+ third-party OSS models (DeepSeek, Qwen, Llama, GLM), so coding quality depends on the chosen model and SWE-bench is N/A (0, none published by the platform). Enterprise/private deployment via Reserved/Dedicated GPUs and BYOC (self-host into your own cloud); no published SSO/SAML, air-gap, or SOC2/ISO attestation ⚠ verify. Data posture: markets \"zero data retention\" / E2E-encrypted neutral processor — treat as no-train, but no explicit contractual ZDR/code-no-train clause independently confirmed ⚠ verify for regulated workloads. Singapore-registered global entity (siliconflow.com) with mainland-CN-operated infra (siliconflow.cn); HK-reachable without VPN/US tenant. Investors include China Development Bank Capital & China Unicom — note CN ownership exposure for a CITIC-group buyer.","tiers":[{"label":"Pay-as-you-go (Serverless)","unit":"paygo","key":"payg","limit":"Per-token / per-image / per-second billing across 200+ models; LLMs ~$0.04-$1.6/M input, $0.18-$4.4/M output; images $0.014-$0.08; video $0.29; no monthly fee, no minimum commitment; $1 free credits to start","monthly":0},{"label":"Reserved / Dedicated GPUs","unit":"infra","key":"reserved","limit":"Guaranteed reserved GPU capacity & dedicated endpoints; custom/contact-sales pricing (infra-priced, not per-seat) ⚠ verify","monthly":0},{"label":"Enterprise / BYOC","unit":"infra","key":"enterprise","limit":"Bring-Your-Own-Cloud deployment + volume discounts; custom pricing via sales ⚠ verify","monthly":0}],"trains_on_code":"no","hk_ok":true,"hosting":"Singapore-registered, CN infra","trains_scope":"Model-hosting proxy/gateway for 200+ third-party OSS models (DeepSeek, Qwen, Llama, GLM); hosts no proprietary model of its own and does not train on user code. Markets a \"zero data retention\" / end-to-end-encrypted posture and acts as a neutral processor handling interaction data per user instructions. Privacy policy retention is \"as long as necessary\" and is silent on an explicit code-training clause, so no formal contractual ZDR/opt-out is independently published. ⚠ verify retention terms for regulated use.","billing_note":"Pure PAYG: per-token (LLM ~$0.04-$1.6/M input, $0.18-$4.4/M output), per-image ($0.014-$0.08), per-video (~$0.29), per-1k-char / per-minute audio. No per-seat subscription. $1 free credits to start; no minimum commitment; monthly spend limits configurable. Volume discounts for high usage via sales. Reserved-GPU and BYOC tiers are infra/custom-priced (contact sales). ⚠ verify exact custom-tier rates.","coding_quality":3,"confidence":"medium","link_status":"live","swe_bench":0},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":false,"soc2":true},"name":"GitHub Copilot","link":"https://github.com/features/copilot/plans","sources":["https://github.com/features/copilot/plans","https://github.blog/news-insights/company-news/github-copilot-is-moving-to-usage-based-billing/","https://github.blog/news-insights/company-news/updates-to-github-copilot-interaction-data-usage-policy/"],"category":"ide_agent","endpoint":"native","region":"global","slug":"github-copilot","model":"GPT-5 / Claude Opus 4.x / Gemini 2.5 (selectable)","note":"Plans (per seat/mo) all confirmed live on github.com/features/copilot/plans (June 2026): Free $0, Pro $10, Pro+ $39, Max $100, Business $19, Enterprise $39. Cheapest paid = Pro $10; top individual tier = Max $100. June 2026 shift to usage-based GitHub AI Credits (per-token): completions stay free, only chat/agent/CLI/code-review draw credits. Live page now lists included credits as Pro $15 / Pro+ $70 / Max $200 (seed's '~$70 Pro+ AI Credits' confirmed). Business/Enterprise promo credits run 1 Jun–1 Sep 2026 (consistent with 'thru Aug 2026'). Training (per 24 Apr 2026 policy, confirmed on github.blog): Business & Enterprise NOT trained; Free/Pro/Pro+ interaction data (incl code snippets) trained by DEFAULT unless opted out — private repo code AT REST never used. trains_on_code='no' (prompt/interaction data, opt-out, not repo code at rest). Microsoft/GitHub-served → HK-usable without VPN or US tenant; for code confidentiality use Business/Enterprise (no-train, SSO/SCIM, SOC 2). SWE-bench: no official GitHub-published score (model-dependent) ⚠ verify.","tiers":[{"label":"Free","unit":"seat","key":"free","limit":"2,000 completions/mo + limited chat/agent; multi-model; CLI. Interaction data trained on by default unless opt-out (from 24 Apr 2026)","monthly":0},{"label":"Pro","unit":"seat","key":"pro","limit":"Unlimited completions; cloud agent; $15/mo AI Credits for chat/agent/CLI","monthly":10},{"label":"Pro+","unit":"seat","key":"pro-plus","limit":"Premium models incl Opus; $70/mo AI Credits","monthly":39},{"label":"Max","unit":"seat","key":"max","limit":"Individual top tier; $200/mo AI Credits; priority new-model access","monthly":100},{"label":"Business","unit":"seat","key":"business","limit":"Org admin/policy; $19/mo AI Credits (3,000 promo credits/user thru 1 Sep 2026); no-train; pooled credits","monthly":19},{"label":"Enterprise","unit":"seat","key":"enterprise","limit":"Org-wide + knowledge base; $39/mo AI Credits (7,000 promo credits/user thru 1 Sep 2026); no-train; budget controls","monthly":39}],"trains_on_code":"no","hk_ok":true,"hosting":"Microsoft (Azure/GitHub-hosted, multi-model Claude/GPT/Gemini)","trains_scope":"Business & Enterprise interaction data is contractually NOT used for training. Free/Pro/Pro+: from 24 Apr 2026 interaction data (inputs incl code snippets, accepted/modified outputs, file names, repo structure) is used for training by DEFAULT unless opted out in Privacy settings. Private-repo content AT REST is never used on any plan; active-session private repo code is interaction data and may be trained on (Free/Pro/Pro+) unless opted out.","billing_note":"Per-seat monthly USD shown. June 1 2026: Premium Request Units replaced by usage-based GitHub AI Credits (per-token, input+output+cached at API rates). Completions & Next Edit suggestions FREE on all plans, no credits consumed; chat/agent/code review/CLI draw credits. Included monthly AI Credits confirmed live: Pro $15, Pro+ $70, Max $200, Business $19, Enterprise $39. Business & Enterprise get promotional enhanced credits 1 Jun–1 Sep 2026 (Business 3,000/user vs std 1,900; Enterprise 7,000/user vs std 3,900); credits pooled org-wide with admin budget caps. Annual Pro/Pro+ subscribers stay on legacy premium-request billing until expiry. No published min-seat for Business. Verified students free via Student Pack. No standalone annual % discount published for new plans ⚠ verify.","coding_quality":4.5,"confidence":"high","link_status":"live","swe_bench":0},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":false,"soc2":true},"name":"Cursor","link":"https://cursor.com/pricing","sources":["https://cursor.com/pricing","https://cursor.com/security","https://dev.to/rahulxsingh/cursor-pricing-in-2026-hobby-pro-pro-ultra-teams-and-enterprise-plans-explained-4b89","https://aiproductivity.ai/blog/cursor-pricing/","https://www.nocode.mba/articles/cursor-pricing"],"category":"ide_agent","endpoint":"native","region":"US","slug":"cursor","model":"Claude/GPT/Gemini + Composer (selectable)","note":"Cursor (Anysphere) IDE agent. Tiers reshaped vs our seed: \"Business $40\" is now \"Teams\" with two seat types — Standard $40 and Premium $120; individual ladder Hobby $0 / Pro $20 / Pro+ $60 / Ultra $200 unchanged. Composer + Claude/GPT/Gemini frontier models. No training on code with Privacy Mode (default Pro/Business, admin-enforceable). SOC 2 Type II (on request via trust.cursor.com), SAML/OIDC SSO + SCIM (Enterprise). No self-host / on-prem / air-gap — cloud-only (US, AWS). Cursor explicitly states it maintains NO infrastructure in China, and agent/cloud features are US-cloud-locked → NOT cleanly HK-usable for a CITIC-group (mainland-owned) firm without a US tenant; treat as a compliance flag. No official SWE-bench Verified figure published ⚠ verify. Premium-seat $120 and annual discounts from 2026 third-party sources ⚠ verify.","tiers":[{"label":"Hobby (Free)","unit":"seat","key":"hobby","limit":"Free; limited Agent requests + limited Tab completions, no card required","monthly":0},{"label":"Pro","unit":"seat","key":"pro","limit":"Extended agent limits, frontier models, MCPs/skills/hooks, cloud agents, Bugbot; ~$20 usage credit then usage-based","monthly":20},{"label":"Pro+","unit":"seat","key":"pro-plus","limit":"~3x Pro usage credits","monthly":60},{"label":"Ultra","unit":"seat","key":"ultra","limit":"~20x Pro usage, priority feature access","monthly":200},{"label":"Teams (Standard)","unit":"seat","key":"teams-standard","limit":"Per-seat usage pool + admin: centralized billing, SAML/OIDC SSO, team privacy mode, analytics, agentic code review","monthly":40},{"label":"Teams (Premium)","unit":"seat","key":"teams-premium","limit":"~5x Standard usage per seat","monthly":120},{"label":"Enterprise","unit":"seat","key":"enterprise","limit":"Custom: pooled usage, SCIM, access controls, audit logs, AI code tracking API, priority support (contact sales)","monthly":0}],"trains_on_code":"no","hk_ok":false,"hosting":"Anysphere (Cursor cloud, US backend)","trains_scope":"Privacy Mode (default-on for Pro/Business, enforceable team-wide by admin) guarantees code is not used for training by Cursor or its model providers, via technical + contractual controls. Without Privacy Mode some data may be retained to improve product. ZDR available on request/enterprise.","billing_note":"Annual billing ~20% off all paid plans (Pro ~$16/mo, Teams Standard ~$32/mo, Premium ~$96/mo billed annually). Sold only direct via cursor.com (no resellers). Pro+/Ultra/Premium-seat prices from 2026 third-party sources ⚠ verify.","coding_quality":4.5,"confidence":"high","link_status":"live","swe_bench":0},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":true,"soc2":true},"name":"Windsurf / Devin Desktop (Cognition)","link":"https://devin.ai/pricing","sources":["https://devin.ai/pricing","https://devin.ai/blog/windsurf-pricing-plans/","https://devin.ai/security","https://www.cloudzero.com/blog/windsurf-pricing/","https://www.digitalapplied.com/blog/windsurf-becomes-devin-desktop-ide-migration-2026","https://cognition.ai/blog/swe-1-6","https://www.reco.ai/learn/windsurf-security","https://docs.devin.ai/desktop/devin-desktop-faq"],"category":"ide_agent","endpoint":"native","region":"US","slug":"windsurf-devin-desktop","model":"SWE-1 / Claude / GPT (selectable) + Cascade agent","note":"Windsurf rebranded to Devin Desktop (Cognition) on 2 Jun 2026; windsurf.com/pricing now 308-redirects to the live devin.ai/pricing. Plans verified: Free $0 / Pro $20 / Max $200 / Teams $80 base + $40 per dev seat / Enterprise custom — same pricing post-rebrand. Cascade agent (SWE-1.6, plus Claude/GPT/Gemini) being replaced by Rust 'Devin Local'; Cascade supported only through 1 Jul 2026. Code is processed ephemerally and not used to train models; ZDR default on Team/Enterprise; SOC 2 Type II + FedRAMP High (via Palantir FedStart) ⚠ verify; SSO/SAML/SCIM/RBAC and VPC/self-host on Enterprise. No fully air-gapped/offline deploy attested ⚠ verify. Cognition reports SWE-1.6 on SWE-Bench Pro, not SWE-bench Verified, so no current Verified % (set 0). HK risk: Cognition is a US-cloud-locked agent product (Devin Cloud/Desktop, Windows VM infra) — not cleanly usable from Hong Kong without a US tenant; weak fit for a CITIC-group / mainland-owned broker. coding_quality 4 (strong IDE agent) ⚠ verify.","tiers":[{"label":"Free","unit":"seat","key":"free","limit":"Light agent quota; limited model selection; unlimited Tab/inline edits; SWE-1.6 free tier (200 tok/s)","monthly":0},{"label":"Pro","unit":"seat","key":"pro","limit":"Frontier models (Claude/OpenAI/Gemini) + free SWE-1.6 fast (950 tok/s); ~7-190 msgs/day by model tier; daily/weekly quotas","monthly":20},{"label":"Max","unit":"seat","key":"max","limit":"Everything in Pro with much higher quotas (~42-1,190 msgs/day by model tier)","monthly":200},{"label":"Teams","unit":"seat","key":"teams","limit":"$80/mo base + $40/mo per full dev seat; per-seat usage spans Devin Cloud/Desktop/CLI/Review; org collaboration","monthly":40},{"label":"Enterprise","unit":"seat","key":"enterprise","limit":"Custom pricing; SAML/OIDC SSO, SCIM, RBAC, VPC/self-host deployment, custom terms & support","monthly":0}],"trains_on_code":"no","hk_ok":false,"hosting":"Cognition (US cloud)","trains_scope":"Customer code processed ephemerally and not used to train underlying models; zero data retention is the default on Team/Enterprise tiers and code-storing features require explicit opt-in. ⚠ verify exact consumer-tier ZDR scope.","billing_note":"No public annual discount or minimum-seat requirement. Teams = $80/mo base + $40/mo per additional full developer seat (per-seat usage pools across all Devin products). Credit system retired Mar 2026 for daily/weekly quotas; existing Pro/Teams subs grandfathered. Enterprise custom-quoted. USD.","coding_quality":4,"confidence":"high","link_status":"live","swe_bench":0},{"flags":{"enterprise":true,"sso":false,"air_gap":false,"self_host":false,"soc2":false},"name":"Trae (ByteDance / BytePlus)","link":"https://www.trae.ai/pricing","sources":["https://www.trae.ai/pricing","https://www.trae.ai/privacy-policy","https://hokai.io/hub/tools/trae","https://vibecoding.app/blog/trae-review","https://www.theregister.com/2025/07/28/bytedance_trae_telemetry/","https://blog.unit221b.com/dont-read-this-blog/unveiling-trae-bytedances-ai-ide-and-its-extensive-data-collection-system","https://cybernews.com/security/bytedance-ai-coding-tool-trae-data-collection/"],"category":"ide_agent","endpoint":"native","region":"CN","slug":"trae","model":"Claude 4 / GPT / DeepSeek R1 (selectable) + SOLO agent","note":"Official pricing page live (trae.ai/pricing) and confirms 5 tiers: Free $0 / Lite $3 / Pro $10 / Pro+ $30 / Ultra $100, all per-seat/mo; annual saves 25%; Pro 7-day trial; PAYG for overages. SOLO agent + Claude/GPT/DeepSeek models. ByteDance IDE — usable from HK (mainland-CN/global endpoints, no US tenant needed) and acceptable for a CITIC-group firm. KEY RISK for a regulated broker: ByteDance ownership + documented heavy telemetry, ~5yr retention, no reliable opt-out (The Register, Unit221b, Cybernews); no SSO/SAML, no self-host/air-gap, no SOC2/ISO. Trae states it does NOT train models on user code, but parent-company data handling is the real concern. Enterprise inquiries handled via BytePlus sales (no org admin/policy controls published) ⚠ verify. No published SWE-bench Verified score.","tiers":[{"label":"Free","unit":"seat","key":"free","limit":"5,000 autocompletions/mo; limited basic usage; 2 concurrent cloud tasks; limited SOLO mode","monthly":0},{"label":"Lite","unit":"seat","key":"lite","limit":"Everything in Free plus more basic usage; 5,000 autocompletions; 2 concurrent cloud tasks","monthly":3},{"label":"Pro","unit":"seat","key":"pro","limit":"Unlimited autocomplete; premium-model fast requests; SOLO mode; 7-day free trial","monthly":10},{"label":"Pro+","unit":"seat","key":"pro-plus","limit":"~3.5x Pro usage; more concurrent cloud tasks","monthly":30},{"label":"Ultra","unit":"seat","key":"ultra","limit":"~20x Pro usage; model early access; up to 20 concurrent cloud tasks","monthly":100},{"label":"Enterprise (assumed)","unit":"seat","key":"enterprise","limit":"Org admin · SSO · centralised billing via BytePlus sales — no public price; $75/seat planning assumption ⚠ verify","monthly":75}],"trains_on_code":"no","hk_ok":true,"hosting":"ByteDance / BytePlus","trains_scope":"Trae officially states it does not train AI models on user data/code. Caveat: ByteDance telemetry is extensive (file paths, hardware IDs), ~5yr retention, no clean opt-out per The Register/Unit221b reports — privacy risk, not model-training on code. ⚠ verify","billing_note":"Annual billing saves 25%. Pro has a 7-day free trial then $10/mo. PAYG/on-demand available for overages. No published min-seats. Enterprise routed to BytePlus sales (no public price).","coding_quality":3.5,"confidence":"high","link_status":"live","swe_bench":0},{"flags":{"enterprise":true,"sso":false,"air_gap":false,"self_host":false,"soc2":false},"name":"Zed","link":"https://zed.dev/pricing","sources":["https://zed.dev/pricing","https://zed.dev/docs/account/plans-and-pricing","https://zed.dev/blog/zed-for-business","https://zed.dev/docs/ai/privacy-and-security","https://zed.dev/docs/ai/ai-improvement"],"category":"ide_agent","endpoint":"native","region":"US","slug":"zed","model":"BYOK any (Claude/GPT/Gemini/local Ollama)","note":"Open-source Rust editor; pricing confirmed live on official page (Free $0 / Pro $10 / Business $30/seat), matches seed unchanged. Strong privacy posture: no-train by default + ZDR with hosted providers; Business adds org-wide policy/RBAC/spend caps. SSO/SAML/SCIM and SOC 2 are roadmap, NOT yet shipped/attested — blockers for a regulated CITIC-group firm requiring SSO+SOC2 before approval. No self-host/on-prem/air-gap. BYOK (Anthropic/OpenAI/Google/Bedrock/Mistral/Deepseek/etc.) + Ollama/local at no extra charge. HK-usable: desktop app, hosted AI via standard global providers, BYOK works without US tenant or VPN. No published SWE-bench (relies on chosen model). coding_quality ⚠ verify (model-dependent).","tiers":[{"label":"Personal (Free)","unit":"seat","key":"free","limit":"Free forever; limited edit predictions; no Zed-hosted AI but BYOK any model + Ollama / local OpenAI-compatible; external agents (Claude/Codex CLI)","monthly":0},{"label":"Pro","unit":"seat","key":"pro","limit":"Unlimited edit predictions + $5/mo hosted-token credit; usage-based beyond at provider list +10%; 2-wk trial w/ $20 credit (Opus excluded)","monthly":10},{"label":"Business","unit":"seat","key":"business","limit":"Org-wide AI policy/data-governance, RBAC, unified spend caps; prompt-sharing & edit-prediction training off by default (members can't override); BYOK at no extra charge; no included AI credits","monthly":30}],"trains_on_code":"no","hk_ok":true,"hosting":"Zed Industries (BYOK or Zed-hosted models)","trains_scope":"Does not store prompts or train on your code on any plan unless you explicitly opt in / share feedback; code+prompts discarded after each request. Zed-hosted models run under zero-data-retention agreements with Anthropic/Google/OpenAI/xAI (except provider safety-retention models). Business: prompt-sharing & edit-prediction training off by default org-wide, non-overridable.","billing_note":"Self-serve, no minimum seats; order forms available at 25+ seats. No annual-discount published (monthly billing; Business cancel monthly). Student plan: Pro free for 1yr + $10/mo token credit (Opus excluded). May 2026 restructure: Pro cut to $10, new Business $30/seat replaced Teams. USD.","coding_quality":4,"confidence":"high","link_status":"live","swe_bench":0},{"flags":{"enterprise":true,"sso":true,"air_gap":true,"self_host":true,"soc2":true},"name":"JetBrains AI (Junie)","link":"https://www.jetbrains.com/ai-ides/buy/","sources":["https://www.jetbrains.com/ai-ides/buy/","https://lp.jetbrains.com/ai-ides-faq/","https://www.jetbrains.com/legal/docs/terms/jetbrains-ai-service/","https://aiproductivity.ai/pricing/jetbrains-ai-assistant/","https://devtoolsreview.com/pricing/jetbrains-ai-pricing/","https://youtrack.jetbrains.com/articles/SUPPORT-A-661/AI-Assistant-security-certifications","https://blog.jetbrains.com/junie/2026/06/junie-coding-agent-out-of-beta/"],"category":"ide_agent","endpoint":"native","region":"EU","slug":"jetbrains-ai-junie","model":"Claude / GPT / Gemini (selectable) + Junie agent","note":"Pricing confirmed across multiple 2026 sources: AI Free $0 / AI Pro $10 (org $20/seat) / AI Ultimate $30 (org $60/seat) / AI Enterprise custom — the seed missed the org-tier doubling and the Enterprise tier. Junie agent + Claude/GPT/Gemini + local models (Ollama/LM Studio). No-train by default with zero data retention from third-party LLM providers; training only on explicit opt-in (own Mellum model). AI Enterprise adds on-prem/air-gapped LLM, SAML/SCIM SSO, SOC 2 Type II via JetBrains IDE Services/Hub. Requires a separate paid IDE subscription. Junie ~60.8% SWE-bench Verified (out of beta Jun 2026). HK-reachable (EU/Czech-HQ global vendor; supports local/on-prem models, no US-tenant lock). Official /ai-ides/buy price table is JS-rendered so the fetcher couldn't read it directly — figures corroborated by independent 2026 sources ⚠ verify exact AI-credit counts and Enterprise $ on the live page.","tiers":[{"label":"AI Free","unit":"seat","key":"free","limit":"$0; 3 AI credits / 30 days for cloud models, unlimited code completion + local models (Ollama/LM Studio)","monthly":0},{"label":"AI Pro (individual)","unit":"seat","key":"pro","limit":"$10/mo ($20/seat for orgs); ~10 AI credits/mo, Junie agent + Claude/GPT/Gemini, multi-model switching","monthly":10},{"label":"AI Ultimate (individual)","unit":"seat","key":"ultimate","limit":"$30/mo ($60/seat for orgs); ~35 AI credits/mo (+$5 bonus), priority model access, advanced agent","monthly":30},{"label":"AI Enterprise","unit":"seat","key":"enterprise","limit":"Custom pricing; on-prem/air-gapped LLM via IDE Services, org admin, SAML/SCIM, zero data retention ⚠ verify exact $","monthly":0}],"trains_on_code":"no","hk_ok":true,"hosting":"JetBrains (cloud, multi-model)","trains_scope":"No-train by default. ToS: JetBrains will not use Your Inputs/Data/Outputs to train any models, and AI subcontractors (OpenAI/Anthropic/Google) commit not to train on them (zero data retention for inference). Training on user content only if you explicitly opt in (used to improve JetBrains' own Mellum, never shared). AI Enterprise = guaranteed zero data retention + on-prem option.","billing_note":"Per-seat monthly. Individual: AI Pro $10, AI Ultimate $30. Organization/business: AI Pro $20/seat, AI Ultimate $60/seat. AI Enterprise = custom quote (on-prem/air-gap). Separate from IDE license — needs a JetBrains IDE subscription too. AI Credits metered; unused quota does not roll over; top-up credits valid 12 months. No annual discount published ⚠ verify.","coding_quality":4,"confidence":"high","link_status":"live","swe_bench":60.8},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":false,"soc2":true},"name":"Replit","link":"https://replit.com/pricing","sources":["https://replit.com/pricing","https://replit.com/commercial-agreement","https://docs.replit.com/references/integrations/replit-ai-integrations","https://aitoolpick.org/blog/replit-pricing-2026/","https://pecollective.com/tools/replit-pricing/","https://costbench.com/software/developer-tools/replit/"],"category":"ide_agent","endpoint":"native","region":"US","slug":"replit","model":"Claude (Replit Agent, hosted)","note":"US cloud IDE + Agent (runs on Anthropic Claude). Live pricing confirmed: Starter $0 / Core $20 ($25 credits) / Pro $100 flat up-to-15-builders (Pro replaced Teams Feb 2026) / Enterprise custom. SOC 2 Type 2 attested; Enterprise adds SSO/SAML, single-tenant, VPC peering, ZDR. No on-prem/air-gap (US-hosted SaaS only). Does NOT train base models on customer code (private by default; tuning only on request for customer's sole use). HK-reachable (US SaaS, no US-tenant requirement) — but US-cloud data residency may concern a CITIC-group firm ⚠ verify region/residency for sensitive repos. SWE-bench: none published. Coding quality is app-build/prototyping oriented rather than top-tier autonomous SWE.","tiers":[{"label":"Starter (Free)","unit":"seat","key":"starter","limit":"Free daily Agent credits; built-in DB & design tools; publish up to 1 app","monthly":0},{"label":"Replit Core","unit":"seat","key":"core","limit":"$25/mo credits, up to 5 collaborators, 2 parallel agents, unlimited workspaces","monthly":20},{"label":"Replit Pro","unit":"seat","key":"pro","limit":"Flat for up to 15 builders + 50 viewers; $100/mo credits, 10 parallel agents, 28-day DB rollbacks, premium support","monthly":100},{"label":"Enterprise","unit":"seat","key":"enterprise","limit":"Custom pricing; SSO/SAML, single-tenant, VPC peering, ZDR endpoints, region selection, dedicated support","monthly":0}],"trains_on_code":"no","hk_ok":true,"hosting":"Replit (cloud IDE, Agent)","trains_scope":"Customer Content private by default, not used to train base models; tuning only on customer request for own sole use. Enterprise ZDR endpoints + BYOK. Free-tier prompts may be used by underlying model providers.","billing_note":"Annual billing ~20% off: Core $16/mo ($240/yr eff. $20 vs $25 monthly), Pro ~$95/mo annual ($1,140/yr) vs $100 monthly flat. Pro replaced the old Teams plan (launched Feb 2026), flat for up to 15 builders (~under $7/builder). Effort-based pay-as-you-go credits beyond plan allowance; no default spend cap. Enterprise custom-priced.","coding_quality":3.5,"confidence":"high","link_status":"live","swe_bench":0},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":false,"soc2":true},"name":"Google Gemini Code Assist","link":"https://cloud.google.com/products/gemini/pricing","sources":["https://cloud.google.com/products/gemini/pricing","https://docs.cloud.google.com/gemini/docs/codeassist/overview","https://docs.cloud.google.com/gemini/docs/discover/data-governance","https://docs.cloud.google.com/gemini/docs/codeassist/gemini-3","https://developers.google.com/gemini-code-assist/resources/release-notes","https://aiproductivity.ai/pricing/gemini-code-assist/","https://www.finout.io/blog/gemini-pricing-in-2026","https://docs.cloud.google.com/gemini-enterprise-agent-platform/resources/zero-data-retention"],"category":"ide_agent","endpoint":"native","region":"global","slug":"gemini-code-assist","model":"Gemini 2.5 Pro","note":"Two paid tiers only as of Jun 2026: Standard $19 / Enterprise $45 per seat/mo on ANNUAL commit (monthly commit $22.80 / $54). Seed's $19/$45 = annual rates (correct); monthly is higher. Free/individual tier ended 18 Jun 2026. Model advanced beyond seed: baseline Gemini 2.5, with Gemini 3 (3.1 Pro / 3.0 Flash) now available to Standard/Enterprise via Preview release channel — not just \"2.5 Pro.\" No training on customer code; ZDR-equivalent for enterprise via Vertex/DPA. Served from Google Cloud (Vertex AI) — HK-reachable without US tenant/VPN, suitable for a CITIC-group firm. Google Cloud carries SOC 2 / ISO 27001. No on-prem/air-gap. No official SWE-bench Verified score published for Code Assist ⚠ verify.","tiers":[{"label":"Standard","unit":"seat","key":"standard","limit":"Agent mode, Gemini CLI, code completion/chat, Google Cloud (Vertex) integration; baseline Gemini 2.5, Gemini 3 in Preview channel. $19/seat annual; $22.80/seat monthly commitment","monthly":19},{"label":"Enterprise","unit":"seat","key":"enterprise","limit":"Standard + org admin/policy controls, private code repo context, VPC-SC/IAM, expanded cloud tools. $45/seat annual; $54/seat monthly commitment; 30-day free trial up to 50 users","monthly":45}],"trains_on_code":"no","hk_ok":true,"hosting":"Google Cloud (Vertex/Gemini)","trains_scope":"Standard & Enterprise do not use customer prompts, code, or responses to train Gemini models (\"Gemini doesn't use your prompts or its responses as data to train its models\"). Data encrypted in-transit; ZDR-equivalent terms available to eligible enterprise customers via Vertex AI / Cloud DPA amendment.","billing_note":"Standard $19/seat/mo annual vs $22.80/mo monthly commit; Enterprise $45/seat/mo annual vs $54/mo monthly commit (~17% annual discount). Enterprise 30-day free trial up to 50 users. Free/individual tier DISCONTINUED 18 Jun 2026. USD; billed via Google Cloud.","coding_quality":4,"confidence":"high","link_status":"live","swe_bench":0},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":false,"soc2":true},"name":"Google Antigravity","link":"https://antigravity.google/pricing","sources":["https://antigravity.google/pricing","https://blog.google/products-and-platforms/products/google-one/google-ai-subscriptions/","https://antigravity.google/terms","https://aicodingtools.im/pricing","https://aitooltier.com/pricing/antigravity","https://www.toolworthy.ai/tool/google-antigravity","https://discuss.ai.google.dev/t/antigravity-data-training-opt-out/125236","https://developers.google.com/gemini-code-assist/resources/privacy-notice-gemini-code-assist-individuals"],"category":"ide_agent","endpoint":"native","region":"global","slug":"google-antigravity","model":"Gemini 3 (multi-agent)","note":"Gemini-family agent-first multi-agent IDE. Tiers re-confirmed vs official Google I/O 2026 blog (blog.google/Google One): Free $0 / AI Pro $20 (cheapest paid) / AI Ultra mid $100 / AI Ultra top $200 (officially cut from $250 at I/O, May 2026) + PAYG credits $0.01 ($199/20k). Top tier is officially named \"AI Ultra\" — third-party sites call it \"AI Ultra Max\"; kept official label. Official pricing page is JS-rendered and returns no readable content via fetch (link_status=unreadable; canonical URL believed correct, corroborated by aggregators ⚠ verify live page). TRAINING DOWNGRADED no→unclear: per Antigravity Additional ToS, consumer/default tiers (Free/Pro/Ultra) use interaction data — including agent code/repo context — to improve Google ML models BY DEFAULT, with opt-out; no-train only via Workspace/GCP. The priced seat tiers therefore train-by-default unless opted out — material for a HK financial firm; route via GCP/Gemini Enterprise for no-train, in-tenant data. Enterprise SSO/SOC 2/audit logs via Gemini Enterprise; no self-host/air-gap (Google-cloud-served). HK-usable without US tenant/VPN. SWE-bench 80.6% is Gemini 3.1 Pro Verified ⚠ verify current default model. Enterprise per-seat price not public ⚠ verify.","tiers":[{"label":"Free","unit":"seat","key":"free","limit":"All models (Gemini 3.1 Pro, Claude Sonnet/Opus 4.6, GPT-OSS) at low rate limits; quota refreshes ~every 5h; ~20 agent requests/day, Gemini Flash default; unlimited tab completion","monthly":0},{"label":"AI Pro","unit":"seat","key":"pro","limit":"Baseline paid quota (1x); all frontier models, higher rate limits, priority access; shared Gemini quota, compute-used model resetting ~5h up to weekly caps","monthly":20},{"label":"AI Ultra (mid)","unit":"seat","key":"ultra-mid","limit":"~5x Pro token quota in Antigravity + Gemini app; priority access; includes Google AI Ultra perks (20TB storage, etc.)","monthly":100},{"label":"AI Ultra","unit":"seat","key":"ultra","limit":"~20x Pro token quota; highest limits, extended agent sessions, experimental models; reduced from $250 at I/O May 2026","monthly":200},{"label":"AI Credits (overflow)","unit":"paygo","key":"paygo","limit":"$0.01/credit when included quota runs out; bulk $199 for 20,000 credits; auto-charge or stop-on-empty","monthly":0}],"trains_on_code":"unclear","hk_ok":true,"hosting":"Google (agent-first IDE, Gemini-hosted)","trains_scope":"Per Antigravity Additional Terms of Service (2026), on consumer/default tiers (Free/AI Pro/AI Ultra) interaction data — which for an agent IDE includes code/repo context the agent operates on, plus prompts and metadata — IS used by default to develop/improve Google ML models; users can opt out in settings. No-train ONLY when accessed via Google Workspace or Google Cloud (GCP/Gemini Enterprise), where data stays in-tenant. So priced seat tiers train-by-default (opt-out available); enterprise/Cloud path is no-train. ⚠ verify whether opt-out fully excludes source-code context.","billing_note":"USD per seat per month, billed via Google One AI subscription. Top-tier AI Ultra cut $250→$200 at I/O 2026 (officially confirmed on blog.google); new $100 mid-Ultra added same date. AI credits overflow billed PAYG at $0.01/credit ($199/20k). No published annual discount or minimum-seat requirement. Top tier is officially \"AI Ultra\" $200 (third-party aggregators call it \"AI Ultra Max\" — not an official label). ⚠ verify enterprise per-seat (Gemini Enterprise Agent Platform is consumption/contract-priced, not public).","coding_quality":4.5,"confidence":"medium","link_status":"unreadable","swe_bench":80.6},{"flags":{"enterprise":true,"sso":true,"air_gap":true,"self_host":true,"soc2":false},"name":"Cline","link":"https://cline.bot/pricing","sources":["https://cline.bot/pricing","https://cline.bot/blog/introducing-cline-for-enterprise","https://cline.bot/blog/enterprise-ai-coding-that-uses-your-infrastructure-not-ours","https://docs.cline.bot/enterprise-solutions/overview","https://costbench.com/software/ai-coding-assistants/cline/","https://checkthat.ai/brands/cline/pricing","https://toolradar.com/tools/cline/pricing"],"category":"terminal_agent","endpoint":"anthropic+openai","region":"US","slug":"cline","model":"BYOK (any: Claude, GPT, Gemini, DeepSeek, Qwen, local)","note":"Open-source terminal/IDE agent (VS Code + CLI, JetBrains on paid). Free OSS BYOK $0; Teams $20/seat/mo post-Q1 2026 with 10 seats always free; Enterprise custom. Best harness for CN plans — fully provider-agnostic (Anthropic/GPT/Gemini/DeepSeek/Qwen/local, Bedrock), so a CITIC-group firm can route to mainland-CN or self-hosted endpoints; no US-cloud lock-in. Strong no-train guarantee (code never uploaded/indexed/trained). Enterprise supports SSO/OIDC/SCIM, RBAC, audit logs, VPC, self-hosted (K8s/Helm/IaC) and TRUE air-gapped deploy (CMMC/NIST 800-171 use cases). NB: Cline holds NO direct SOC 2/ISO attestation — it states compliance is 'inherited from the AI provider you select', so SOC 2 must come from the chosen model provider, not Cline. No published SWE-bench (harness, not a model). hk_ok: usable from HK with no US tenant. ⚠ verify Teams GA timing/$20 and inherited-compliance posture before procurement","tiers":[{"label":"Open Source (Free)","unit":"paygo","key":"oss","limit":"Free Apache-2.0 OSS; VS Code ext + CLI; BYOK or Cline provider PAYG (no inference markup); MCP marketplace; client-side, community support","monthly":0},{"label":"Teams","unit":"seat","key":"teams","limit":"$20/user/mo (post-Q1 2026); first 10 seats permanently free; adds JetBrains, centralized billing/admin, team dashboard, RBAC","monthly":20},{"label":"Enterprise","unit":"seat","key":"enterprise","limit":"Custom/contact sales; SSO/OIDC/SCIM, audit logs, RBAC, VPC + self-hosted (K8s/Helm/IaC) + air-gapped, OpenTelemetry, SLA, dedicated support","monthly":0}],"trains_on_code":"no","hk_ok":true,"hosting":"Local VS Code/JetBrains extension; BYOK to any endpoint","trains_scope":"Explicit no-train: code/prompts never uploaded, indexed, cached, or used to train/fine-tune Cline models; client-side execution. Cline hosts no model of its own (BYOK/PAYG harness); model-side training governed by whichever provider the firm points it at — opt for a ZDR/no-train provider endpoint.","billing_note":"Teams free through Q1 2026, then $20/user/mo with first 10 seats permanently free (structural discount for small teams). No annual-discount or minimum-seat figures published. Inference billed separately at-cost (BYOK or Cline provider PAYG, no markup). Enterprise custom-quoted. ⚠ verify exact Teams GA date/$20 figure","coding_quality":4.5,"confidence":"high","link_status":"live","swe_bench":0},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":true,"soc2":false},"name":"Kilo Code","link":"https://kilo.ai/pricing","sources":["https://kilo.ai/pricing","https://kilo.ai/pricing/kilo-pass/","https://kilo.ai/privacy","https://blog.kilo.ai/p/coding-plans-are-live-in-kilo","https://github.com/Kilo-Org/kilocode/discussions/6015","https://skywork.ai/blog/kilo-code-ai-review-2025-vs-code-jetbrains/","https://www.bodegaone.ai/blog/real-cost-of-kilo-code"],"category":"terminal_agent","endpoint":"anthropic+openai","region":"global","slug":"kilo-code","model":"BYOK 500+ models (any OpenAI/Anthropic-compatible)","note":"Pricing page LIVE at kilo.ai/pricing (verified Jun 2026). Seed Pass tiers confirmed accurate: Starter $19 / Pro $49 / Expert $199 (USD/seat), plus free OSS BYOK $0 and PAYG Kilo Gateway (500+ models / 60+ providers — confirmed). NEW since seed: Teams $15/user/mo, Enterprise (custom, SSO/OIDC/SCIM/audit logs/SLA), and KiloClaw cloud agent $55/mo (inference billed separately ⚠ verify currency-of-price). No-train confirmed for code; ZDR exists but Enterprise-gated (individual/Pass ZDR not yet shipped ⚠ verify). OSS extension (VS Code/JetBrains/CLI) — BYOK/local enables self-host/air-gap-style use; no published SOC 2/ISO attestation found ⚠ verify. HK-reachable: not US-tenant-locked, globally served, BYOK to any provider. No SWE-bench Verified score published.","tiers":[{"label":"Free & Open Source (Auto Free / BYOK / Local)","unit":"infra","key":"free-byok","limit":"OSS VS Code / JetBrains / CLI extensions; bring your own keys or use free/local models; no hosted credit required","monthly":0},{"label":"Kilo Gateway (PAYG)","unit":"paygo","key":"gateway","limit":"$0/mo + usage at exact provider rates; 500+ models across 60+ providers","monthly":0},{"label":"Kilo Pass — Starter","unit":"seat","key":"pass-starter","limit":"$19 converts 1:1 to credits; up to ~$26.60/mo total with bonus credits","monthly":19},{"label":"Kilo Pass — Pro","unit":"seat","key":"pass-pro","limit":"$49 → up to ~$68.60/mo total with bonus credits","monthly":49},{"label":"Kilo Pass — Expert","unit":"seat","key":"pass-expert","limit":"$199 → up to ~$278.60/mo total with bonus credits","monthly":199},{"label":"Teams","unit":"seat","key":"teams","limit":"$15/user/mo; usage analytics, AI adoption scoring, shared agent modes, centralized billing; 14-day trial","monthly":15},{"label":"KiloClaw (cloud agent)","unit":"seat","key":"kiloclaw","limit":"$55/mo cloud coding agent; AI inference billed separately; 1-day trial (~$51/mo on 6-mo commit)","monthly":55},{"label":"Enterprise","unit":"seat","key":"enterprise","limit":"Custom pricing; SSO/OIDC/SCIM, audit logs, SLA, dedicated support, ZDR providers","monthly":0}],"trains_on_code":"no","hk_ok":true,"hosting":"Local VS Code extension + CLI; BYOK or optional Kilo Pass credits","trains_scope":"Kilo does not train on user code; BYOK/local keeps code on-device; gateway routes prompts to chosen provider under their terms; ZDR available but Enterprise-only.","billing_note":"Kilo Pass converts subscription 1:1 to credits; bonus credits = 50% welcome bonus month 1 then up to 40% on monthly streak; annual = flat 50% bonus every month (e.g. $228 paid → $342 usable; Expert annual ~$588/yr cited). Bonus credits expire at end of each cycle; base credits carry forward (expire after 1yr). Gateway is PAYG at exact provider rates. KiloClaw $55/mo (~$51/mo on 6-mo prepay, $306). Teams $15/user/mo, 14-day trial. Enterprise custom. No min-seat stated. USD.","coding_quality":4,"confidence":"high","link_status":"live","swe_bench":0},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":false,"soc2":false},"name":"OpenCode","link":"https://opencode.ai/docs/zen/","sources":["https://opencode.ai/","https://opencode.ai/docs/zen/","https://opencode.ai/docs/go/","https://opencode.ai/docs/enterprise/","https://opencode.ai/go","https://github.com/opencode-ai/opencode"],"category":"terminal_agent","endpoint":"anthropic+openai","region":"global","slug":"opencode","model":"BYOK 75+ providers (any OpenAI/Anthropic-compatible)","note":"OSS (MIT) terminal-first agent; core product is free BYOK across 75+ providers — you pay providers directly. Since seed: OpenCode now sells first-party access. Three paid paths: (1) Go = flat $10/mo ($5 first mo) for ~13 curated open/Chinese-lab coding models with dollar caps ($12/5hr, $30/wk, $60/mo); (2) Zen = prepaid PAYG, zero markup, auto-reload; (3) Enterprise = per-seat custom quote with SSO via org IdP and central config to restrict/route providers. No training on user code; OpenCode stores no code (a few free/trial Zen models collect data during feedback — avoidable). Self-host on roadmap (NOT GA ⚠), no air-gap, no published SOC 2/ISO ⚠ verify. HK-reachable: BYOK harness, no US-tenant lock; mainland-CN model endpoints fine for a CITIC-group firm. No SWE-bench Verified score published ⚠. /pricing path 404s — canonical pricing lives under /docs/zen/ and /docs/go/.","tiers":[{"label":"Free (BYOK / OSS)","unit":"paygo","key":"free-byok","limit":"MIT OSS terminal TUI; bring your own key for 75+ providers (Anthropic, OpenAI, Gemini, Bedrock, Copilot, etc.); you pay providers directly; free Zen models also usable","monthly":0},{"label":"OpenCode Go","unit":"seat","key":"go","limit":"Flat monthly access to ~13 curated open/Chinese-lab coding models (GLM-5.x, Kimi K2.x, Qwen3.x, DeepSeek V4, MiniMax M3, MiMo); dollar-value caps: $12/5hr, $30/week, $60/month; $5 first month","monthly":10},{"label":"OpenCode Zen (PAYG)","unit":"paygo","key":"zen","limit":"Prepaid balance, zero markup on curated benchmarked models (free → ~$30 in/$180 out per 1M tok); auto-reload $20 when balance <$5; admin monthly spend caps; team workspaces free during beta","monthly":0},{"label":"Enterprise","unit":"seat","key":"enterprise","limit":"Custom per-seat quote (contact sales); central config to lock providers / route via org gateway; SSO via org IdP; no token charge if you bring your own LLM gateway; self-host on roadmap","monthly":0}],"trains_on_code":"no","hk_ok":true,"hosting":"Local terminal TUI; BYOK to 75+ providers","trains_scope":"OpenCode stores no code/context and runs locally; Zen/Go model providers follow zero-retention and do not train on user data (OpenAI/Anthropic retain 30 days, no training). Exception: a handful of free/trial Zen models collect data during feedback periods — avoidable.","billing_note":"OpenCode Go: $5 first month then $10/mo; usage capped by dollar-equivalent ($12/5hr, $30/wk, $60/mo). Zen is prepaid PAYG with zero markup, auto-reload $20 below $5 balance. Team workspaces free during beta. Enterprise per-seat by custom quote; no token fee if you bring your own LLM gateway. No published annual discount or min-seats. ⚠ verify Enterprise per-seat figure (quote-only).","coding_quality":4,"confidence":"high","link_status":"live","swe_bench":0},{"flags":{"enterprise":false,"sso":false,"air_gap":false,"self_host":false,"soc2":false},"name":"Aider","link":"https://aider.chat","sources":["https://aider.chat","https://aider.chat/docs/legal/privacy.html","https://aider.chat/docs/more/analytics.html","https://www.morphllm.com/best-ai-coding-agents-2026","https://aiagentslist.com/agents/aider","https://llm-stats.com/benchmarks/aider-polyglot"],"category":"terminal_agent","endpoint":"anthropic+openai","region":"global","slug":"aider","model":"BYOK (any: Claude, GPT, Gemini, DeepSeek, local)","note":"Free MIT-licensed terminal pair-programmer; git-first (auto-commits every edit, full reviewable diff history). BYOK — connects to any LLM (Claude/GPT/Gemini/DeepSeek/local), pay providers directly, no markup. No paid tier, no Aider Cloud, no org/admin/SSO/SOC2 product (self-host is moot: it's a local CLI you already run on your own machine, and runs fully offline with local models — effective air-gap if paired with a local LLM). Coding quality strong but depends entirely on chosen model. Aider publishes the Polyglot leaderboard (model benchmark), not a SWE-bench Verified score for the tool itself — SWE-bench % belongs to the underlying models. HK-reachable: no US tenant/login required; only constraint is whichever model API endpoint you point it at. Self-host/air-gap flags left false as product features ⚠ verify if board wants 'effectively yes via local models'.","tiers":[{"label":"Open Source (BYOK)","unit":"paygo","key":"oss","limit":"Free MIT-licensed CLI; bring your own API key — pay model providers (Anthropic/OpenAI/Google/DeepSeek/local) directly at their token rates, zero markup. Light use ~$5-20/mo, heavy ~$50-200+/mo in model API costs.","monthly":0}],"trains_on_code":"na","hk_ok":true,"hosting":"Local terminal pair-programmer; BYOK to any endpoint","trains_scope":"Pure OSS BYOK harness — hosts no model of its own, so does no training. Opt-in PostHog analytics (anonymous UUID, LLM/token counts, command usage, errors) explicitly \"never collects your code, chat messages, keys or personal info\"; disable with --analytics-disable. Any code-training risk lies entirely with the downstream model provider's ToS, not Aider.","billing_note":"No subscription, no seats, no Aider Cloud, no annual/promo. 100% free forever (MIT). Only cost is third-party LLM API tokens billed by the model provider you choose; local models = $0.","coding_quality":4,"confidence":"high","link_status":"live","swe_bench":0},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":true,"soc2":true},"name":"Devin (Cognition)","link":"https://devin.ai/pricing","sources":["https://devin.ai/pricing","https://cognition.com/blog/new-self-serve-plans-for-devin","https://docs.devin.ai/admin/security","https://trust.cognition.ai/"],"category":"terminal_agent","endpoint":"na","region":"US","slug":"devin","model":"Cognition proprietary orchestration (Claude/GPT backed)","note":"Plans verified live on devin.ai/pricing + Cognition self-serve blog (Apr 14 2026): Free $0 / Pro $20 / Max $200 (NEW) / Teams $80 base + $40 per full dev seat / Enterprise custom. Seed's flat \"Teams $80\" was base-only — real Teams is $80+$40/seat; old Core/Team ACU tiers retired. Autonomous cloud coding agent (Cognition). Does NOT train on customer code by default (opt-in only; Enterprise never-train); retention silent/relationship-duration. SOC 2 Type II attested (Mar 2024); ISO 27001 not stated. Enterprise adds SAML/OIDC SSO + VPC/dedicated deployment (no fully air-gapped/on-prem offline). hk_ok=false: US-cloud-locked agent, no CN endpoint and no self-host outside Enterprise VPC — not freely usable from HK for a CITIC-group firm without a US/Enterprise tenant. SWE-bench Verified % not published on pricing/docs ⚠ verify. Coding quality 4/5 (strong autonomous agent) ⚠ verify.","tiers":[{"label":"Free","unit":"seat","key":"free","limit":"Light quota to code with agents; limited model availability; unlimited inline edits + Tab completions","monthly":0},{"label":"Pro","unit":"seat","key":"pro","limit":"Increased quota; full model access (OpenAI/Claude/Gemini + SWE 1.6); Devin cloud agents; API overage purchasable","monthly":20},{"label":"Max","unit":"seat","key":"max","limit":"Everything in Pro plus significantly higher usage quota for heavy individual use","monthly":200},{"label":"Teams","unit":"seat","key":"teams","limit":"$80/mo base + $40/mo per full dev seat; unlimited members/flex seats; central billing, admin dashboard, priority support; usage-based beyond quota","monthly":40},{"label":"Enterprise","unit":"seat","key":"enterprise","limit":"Custom quote (ACU-based billing); SAML/OIDC SSO, central admin, dedicated/VPC deployment, never-train, priority support","monthly":0}],"trains_on_code":"no","hk_ok":false,"hosting":"Cognition US cloud (autonomous agent)","trains_scope":"No training on customer code by default; uses customer data only if you explicitly opt-in via Data Controls. Enterprise: Cognition never trains on customer data. Retention = duration of customer relationship unless otherwise specified.","billing_note":"Teams = $80/mo platform base + $40/mo per full dev seat (flex seats unlimited). No annual discount advertised. No stated min seats. Self-serve overage billed in USD; Enterprise billed in ACUs (custom). Old Core/Team ACU plans retired Apr 14 2026. USD pricing.","coding_quality":4,"confidence":"high","link_status":"live","swe_bench":0},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":true,"soc2":true},"name":"Factory Droid","link":"https://factory.ai/pricing","sources":["https://factory.ai/pricing","https://docs.factory.ai/pricing","https://docs.factory.ai/cli/account/security","https://every.to/vibe-check/vibe-check-i-canceled-two-ai-max-plans-for-factory-s-coding-agent-droid","https://theaiagentindex.com/agents/factory-ai"],"category":"terminal_agent","endpoint":"anthropic+openai","region":"US","slug":"factory-droid","model":"Multi-model (Claude, GPT, Gemini) via Factory orchestration","note":"Pricing confirmed live on factory.ai/pricing and docs.factory.ai/pricing (June 2026): Pro $20 / Plus $100 (~5x) / Max $200 (~10x), all per-seat/mo, no free tier; Teams (custom, up to 150 seats) and Enterprise (custom, unlimited) on request. Usage/token-billed via rolling rate limits + BYOK overage. No-train confirmed (\"Factory never trains on your code\"); SOC 2 Type II + GDPR; SAML 2.0/OIDC SSO + SCIM; ZDR + customer-managed keys; on-prem/private-cloud at Enterprise; no air-gap stated (⚠ verify air-gap via trust.factory.ai). US-cloud-hosted agent (edits run locally but context/diffs traverse Factory's US cloud) — HK reachability for a CITIC-group firm needs a US tenant/legal review, so hk_ok=false (⚠ verify). No official SWE-bench Verified % currently published (swe_bench=0; vendor claims competitive on Claude 4.5 Sonnet, ⚠ verify).","tiers":[{"label":"Pro","unit":"seat","key":"pro","limit":"Individual; Factory app + Droid CLI/SDK, local & cloud agents, base rate limits (5h/weekly/monthly windows); BYOK allowance then token-billed","monthly":20},{"label":"Plus","unit":"seat","key":"plus","limit":"~5x Pro usage; adds Droid Computers (cloud agents); expanded rate limits","monthly":100},{"label":"Max","unit":"seat","key":"max","limit":"~10x Pro usage; early access to new features; expanded rate limits","monthly":200},{"label":"Teams","unit":"seat","key":"teams","limit":"Custom pricing (contact sales); up to 150 seats; SSO + SAML/SCIM; ZDR option; dedicated support","monthly":0},{"label":"Enterprise","unit":"seat","key":"enterprise","limit":"Custom pricing; unlimited seats; dedicated compute; audit logs; on-prem/private-cloud; full admin controls + SLAs","monthly":0}],"trains_on_code":"no","hk_ok":false,"hosting":"Factory cloud + local CLI/desktop; supports BYOK","trains_scope":"Official docs state \"Factory never trains on your code.\" Zero Data Retention (ZDR) mode available on Teams+; customer-managed encryption keys (BYOK). Shell/file edits run locally, only context+diffs sent to Factory's cloud.","billing_note":"No free tier. Token/usage-billed across 5-hour, weekly and monthly rolling rate-limit windows. BYOK allowance included on individual plans, charged after threshold. No public annual discount listed; Teams/Enterprise are custom (contact sales). Active 2026 third-party coupon sites advertise discount codes (unofficial, ⚠ verify). Min-seats not stated.","coding_quality":4,"confidence":"high","link_status":"live","swe_bench":0},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":false,"soc2":true},"name":"Augment Code","link":"https://www.augmentcode.com/pricing","sources":["https://www.augmentcode.com/pricing","https://www.augmentcode.com/security","https://www.augmentcode.com/blog/1-open-source-agent-on-swe-bench-verified-by-combining-claude-3-7-and-o1","https://www.augmentcode.com/blog/auggie-tops-swe-bench-pro","https://venturebeat.com/ai/augment-code-debuts-ai-agent-with-70-win-rate-over-github-copilot-and-record-breaking-swe-bench-score","https://github.com/augmentcode/augment-swebench-agent"],"category":"terminal_agent","endpoint":"anthropic+openai","region":"US","slug":"augment-code","model":"Multi-model via Augment Context Engine (Claude/GPT backed)","note":"Live pricing page (June 2026) shows only two plans: Business = flat $100/mo for the org (up to 50 seats pooled, NOT per-seat) bundling $100 of usage (LLM at provider rates + ~40% service fee + compute) with pay-as-you-go top-ups; and Enterprise = custom pricing, unlimited users, volume annual discounts. Legacy Indie/$20, Standard/$60, Max/$200 individual credit tiers no longer listed — third-party aggregators citing them are stale. No-train confirmed (\"never train on customer proprietary data\"). SOC 2 Type II + ISO 42001 (note: ISO 42001, not 27001); GDPR/CCPA/HIPAA + BAA. Enterprise SSO/OIDC/SCIM. No self-host / VPC / on-prem / air-gap offering — US SaaS cloud only, so unsuitable where on-prem is mandatory. Context Engine indexes repos for Claude/GPT-backed agent (\"Auggie\"). HK: standard US-hosted web SaaS, reachable from Hong Kong without a US tenant or VPN, but data routes to US cloud (no mainland-CN region) — a data-residency/sovereignty concern for a CITIC-group firm ⚠ verify acceptability. SWE-bench Verified 65.4% is a 2025 published figure (Claude 3.7 + o1 ensemble); benchmark now saturated, no fresh Verified number published for the current agent ⚠ verify.","tiers":[{"label":"Business","unit":"infra","key":"business","limit":"Flat $100/mo for the org (NOT per seat); up to 50 seats included; bundle = $100/mo of usage across LLM + Context Engine + compute (LLM at provider rates + 40% service fee), then pay-as-you-go top-ups (expire 12 months)","monthly":100},{"label":"Enterprise","unit":"seat","key":"enterprise","limit":"Custom per-user pricing; unlimited users; custom/multi-region compute; volume-based annual discounts; adds SSO/OIDC/SCIM, CMEK, ISO 42001","monthly":0}],"trains_on_code":"no","hk_ok":true,"hosting":"Augment cloud; IDE + CLI agent","trains_scope":"Explicit no-train: \"We never train on our customer's proprietary data\"; all paid plans exclude AI training on customer data per Commercial ToS, with data-minimization and indemnification. No training on source code.","billing_note":"Business is a flat $100/mo org price (up to 50 seats pooled), not per-seat; $100 covers usage (LLM at provider rates + ~40% service fee + compute), overage via pay-as-you-go top-ups that expire 12 months after purchase. Enterprise: custom per-user pricing with volume-based annual discounts. Older Indie $20 / Standard $60 / Max $200 consumer credit tiers appear discontinued — no longer on the live pricing page (June 2026); only Business + Enterprise shown. ⚠ verify whether legacy individual tiers persist for existing users.","coding_quality":4,"confidence":"high","link_status":"live","swe_bench":65.4},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":false,"soc2":true},"name":"Amp (Sourcegraph)","link":"https://ampcode.com/pricing","sources":["https://ampcode.com/pricing","https://ampcode.com/manual","https://ampcode.com/security","https://ampcode.com/news/amp-free-frontier","https://ampcode.com/news/amp-free-is-ad-free","https://ampcode.com/news/workspace-entitlements","https://www.g2.com/products/amp-code-amp/pricing"],"category":"terminal_agent","endpoint":"anthropic+openai","region":"US","slug":"amp","model":"Multi-model (Claude/GPT backed), zero-markup passthrough","note":"Amp is a usage-based terminal/IDE coding agent — NO per-seat subscription. Individuals & teams pay PAYG with zero markup on provider API pricing ($5 min). Amp Free now gives ~$10/day (~$300/mo) credits, ad-free, running the frontier smart agent (Opus 4.5 + GPT-5/Gemini-3 subagents); free daily allowance being reduced for some users as of May 2026 ⚠ verify exact cap. Enterprise = same PAYG at +50%, one-time $1,000 min purchase; adds SSO/SAML, directory sync, ZDR, admin/policy. Does NOT train on code (Enterprise cannot enable training; ZDR available). SOC 2 Type II certified + annual pentest; ISO 27001 not stated. No self-host/on-prem/air-gap — US cloud (Anthropic/OpenAI/Google providers). Reachable from HK without a US tenant/VPN. No SWE-bench Verified figure published by Amp ⚠ verify. Seed PAYG/zero-markup/$5-min/no-train/US-cloud all CONFIRMED.","tiers":[{"label":"Amp Free","unit":"paygo","key":"free","limit":"~$10 credits/day (~$300/mo) replenished hourly; frontier smart agent (Opus 4.5 + GPT-5/Gemini-3 subagents), ad-free; CLI/programmatic use consumes paid credits only. Daily allowance being reduced for some users (May 2026 notice)","monthly":0},{"label":"Individual / Team Pay-as-you-go","unit":"paygo","key":"payg","limit":"Zero markup on providers' API pricing; $5 minimum credit purchase; credits expire after 1yr inactivity; no per-seat subscription; CLI + IDE, multi-model","monthly":0},{"label":"Enterprise","unit":"paygo","key":"enterprise","limit":"PAYG at +50% vs individual/team rates; one-time $1,000 USD minimum purchase = $1,000 usage credits; adds SSO/SAML + directory sync, ZDR, admin/policy controls, workspace entitlements","monthly":0}],"trains_on_code":"no","hk_ok":true,"hosting":"Sourcegraph cloud; CLI + IDE agent, PAYG","trains_scope":"Neither Amp nor its subprocessors train models on user data (incl. source code) unless explicitly opted in; Enterprise cannot enable training at all. Enterprise offers Zero Data Retention (LLM inputs/outputs not retained/trained; providers may cache tensors up to 24h). Free/standard threads persist until deleted (removed within 30 days).","billing_note":"No per-seat subscription. Individuals/teams: pure PAYG, zero markup on model-provider API costs, $5 min credit purchase. Free tier: ~$10/day (~$300/mo) of credits, hourly reset, ad-free (since Mar 30 2026), frontier free since Jan 8 2026 but daily allowance being trimmed for some users (May 23 2026). Enterprise: +50% on usage, one-time $1,000 USD minimum (= $1,000 credits). No annual-prepay discount or min-seat published. USD.","coding_quality":4.5,"confidence":"high","link_status":"live","swe_bench":0},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":true,"soc2":true},"name":"Warp","link":"https://www.warp.dev/pricing","sources":["https://www.warp.dev/pricing","https://www.warp.dev/blog/warp-new-pricing-flexibility-byok","https://www.warp.dev/privacy","https://docs.warp.dev/support-and-community/plans-and-billing/plans-pricing-refunds/","https://www.warp.dev/blog/introducing-warp-lightspeed","https://ainativedev.io/news/warp-s-200-month-lightspeed-plan-targets-heavy-duty-ai-coders"],"category":"terminal_agent","endpoint":"anthropic+openai","region":"US","slug":"warp","model":"Multi-model (Claude, GPT, Gemini) + BYOK","note":"Warp restructured pricing late-2025: Free $0 (75 credits/mo) / Build $20 (1,500 credits, BYOK) / Max $200 (18,000 credits) / Business $50/user (org-wide ZDR + SAML SSO, ≤50 seats) / Enterprise custom (self-host cloud agents, admin controls). 10% off annual. SOC 2 certified; no-train under ZDR. Legacy Pro/Turbo/Lightspeed retired Dec 2025. AI cloud is US-hosted (Microsoft/Google/Anthropic backends) — reachable from HK without a US tenant or VPN; not US-tenant-locked. No published SWE-bench Verified score. Seed Max corrected: $200 now carries 18,000 credits; Enterprise tier added; annual discounts surfaced.","tiers":[{"label":"Free","unit":"seat","key":"free","limit":"75 AI credits/mo (150/mo for first 2 months); individually-set Zero Data Retention; no BYOK; limited cloud storage","monthly":0},{"label":"Build","unit":"seat","key":"build","limit":"1,500 AI credits/mo; OpenAI/Anthropic/Google models; BYOK; unlimited collaboration; reload credits","monthly":20},{"label":"Max","unit":"seat","key":"max","limit":"18,000 AI credits/mo (12x Build); everything in Build; better effective credit rate","monthly":200},{"label":"Business","unit":"seat","key":"business","limit":"1,500 credits/user; up to 50 seats; org-wide enforced ZDR; SAML SSO; shared/reload-credit discounts","monthly":50},{"label":"Enterprise","unit":"infra","key":"enterprise","limit":"Custom pricing; self-hosted cloud agents on your infra; BYOK; enterprise admin controls; dedicated account manager","monthly":0}],"trains_on_code":"no","hk_ok":true,"hosting":"Warp cloud-backed AI terminal; supports BYOK","trains_scope":"No training on source code/repos. Under ZDR (reload credits, Business/Enterprise) no customer AI data is retained, stored, or used for training by Warp or contracted providers. On Free/Build, console interactions may be collected to improve AI features but are opt-out (Settings>Privacy); data never sold; ZDR org-wide enforceable only on Business/Enterprise.","billing_note":"10% annual discount: Build $18/mo, Max $180/mo, Business $45/user/mo (billed yearly). Free gives 150 credits/mo for first 2 months then 75/mo. Business capped at 50 seats; Enterprise is custom-quote. Reload credits ~50% cheaper than legacy overages, roll over 12 months, volume discounts. Legacy Pro/Turbo/Lightspeed retired; users migrated at first renewal after Dec 1 2025. USD.","coding_quality":4,"confidence":"high","link_status":"live","swe_bench":0},{"flags":{"enterprise":true,"sso":true,"air_gap":true,"self_host":true,"soc2":true},"name":"Qodo","link":"https://www.qodo.ai/pricing/","sources":["https://www.qodo.ai/pricing/","https://www.qodo.ai/blog/qodo-security-our-commitment-to-data-privacy-and-security/","https://docs.qodo.ai/qodo-documentation/qodo-aware/security-and-privacy","https://www.augmentcode.com/guides/7-soc-2-ready-ai-coding-tools-for-enterprise-security","https://dev.to/rahulxsingh/qodo-ai-pricing-free-vs-teams-vs-enterprise-plans-in-2026-2mh5","https://costbench.com/software/ai-code-review/qodo/"],"category":"terminal_agent","endpoint":"anthropic+openai","region":"US","slug":"qodo","model":"Multi-model (Claude/GPT backed)","note":"RE-VERIFIED June 2026. Live page (/pricing/) loads. Agentic PR-review + CLI/IDE coding agent (formerly Codium AI), multi-model. CORRECTION: live page now shows credit-pooled model — Pro Team $30/mo base + $0.012/credit (up to 30 users), NOT strict per-seat; seed's $38/$30 per-seat figures persist only in secondary sources and look stale ⚠ verify. Free tier reframed as a 14-day Trial (unlimited during trial) — no persistent free 'Developer' tier visible on live page ⚠ verify. NO-TRAIN confirmed on current Qodo security pages: never trains on user code; SOC 2 Type II; 48h deletion. Enterprise adds SSO/SAML, audit logs, BYOK, single-tenant/on-prem + air-gap. US cloud, reachable from HK without VPN, no mainland-CN endpoint. SWE-bench Verified 71.2% (mid-2025 Qodo CLI) est ⚠ verify.","tiers":[{"label":"Trial (Free 14-day)","unit":"seat","key":"trial","limit":"Free 14-day trial, no card; unlimited reviews + unlimited credits during trial, unlimited users. (Note: separate free Open Source program for qualified OSS projects.) Live page no longer lists a persistent free 'Developer' tier ⚠ verify","monthly":0},{"label":"Pro Team","unit":"paygo","key":"pro-team","limit":"$30/mo base, no annual commitment; usage-based credits at $0.012/credit (packs 2,500/5,000/20,000 ≈18/36/143 reviews/mo), overage at same rate w/ customer cap; up to 30 users. Live page is credit-pooled, NOT strict per-seat. Secondary 2026 sources still cite $38/seat monthly /$30/seat annual per-seat framing — stale vs live ⚠ verify","monthly":30},{"label":"Enterprise","unit":"infra","key":"enterprise","limit":"Custom annual quote; 30+ users; SSO/SAML, audit logs, BYOK, single-tenant/on-prem/air-gap, Gerrit, cross-repo, SLA + dedicated CSM. No public per-seat figure ⚠ verify","monthly":0}],"trains_on_code":"no","hk_ok":true,"hosting":"Qodo cloud; IDE + CLI + PR review agent","trains_scope":"Does not train models on user source code/repos; paid-subscriber data never used for training; Context Engine code analyzed then discarded; Teams/Enterprise data deleted from storage within 48h.","billing_note":"Live pricing page (re-fetched June 2026) presents Pro Team as $30/mo base + pooled usage credits ($0.012/credit; packs 2,500/5,000/20,000), up to 30 users, no annual commitment, standard support. NO annual-discount or strict per-seat option shown on the live page — the $38/seat monthly vs $30/seat annual (~21% annual discount) framing comes only from 2026 secondary sources and appears stale ⚠ verify. 14-day free trial, no card, unlimited during trial. Enterprise = custom annual contract, 30+ users, negotiated. SWE-bench 71.2% est ⚠ verify. USD.","coding_quality":4,"confidence":"medium","link_status":"live","swe_bench":71.2},{"flags":{"enterprise":true,"sso":true,"air_gap":true,"self_host":true,"soc2":true},"name":"Tabnine","link":"https://www.tabnine.com/pricing/","sources":["https://www.tabnine.com/pricing/","https://devtoolsreview.com/pricing/tabnine-pricing/","https://www.saasworthy.com/product/tabnine/pricing"],"category":"terminal_agent","endpoint":"anthropic+openai","region":"global","slug":"tabnine","model":"Multi-model + private/air-gapped models","note":"Official pricing page (https://www.tabnine.com/pricing/) is LIVE and confirms the seed: Code Assistant $39/user/mo and Agentic $59/user/mo, both annual billing. Deployment SaaS/VPC/on-prem/fully air-gapped; SSO; SOC 2 + ISO 27001 + GDPR; zero code retention / no training on customer code. Enterprise tier is custom-quoted (price not published ⚠ verify). No free tier (retired 2024). Strong fit for a privacy/compliance-sensitive CITIC-CLSA deployment: fully air-gapped on-prem option means no external dependency — usable from Hong Kong without US tenant or VPN. Microsoft/Google-style global SaaS reachability for cloud tier; self-host removes any geo concern. No SWE-bench Verified score published by Tabnine ⚠ verify. coding_quality is a conservative estimate (model-agnostic; supports BYO frontier LLMs).","tiers":[{"label":"Code Assistant Platform","unit":"seat","key":"code","limit":"AI code completions + IDE chat; SaaS/VPC/on-prem/air-gapped; zero code retention; SOC2/ISO27001/GDPR; annual billing","monthly":39},{"label":"Agentic Platform","unit":"seat","key":"agentic","limit":"Everything in Code Assistant + autonomous agents, MCP tools, Tabnine CLI, Context Engine w/ unlimited codebase connections; annual billing","monthly":59},{"label":"Enterprise (custom quote)","unit":"seat","key":"enterprise","limit":"Custom/'Get a quote'; private deployment, dedicated CSM, expanded IP indemnification; price not published","monthly":0}],"trains_on_code":"no","hk_ok":true,"hosting":"Tabnine cloud, VPC, or air-gapped on-prem","trains_scope":"Privacy-first: \"Zero code retention... no training on your code, no sharing with third parties.\" Tabnine's base models are trained only on permissively licensed open-source. Optional org-private custom training on the customer's own repos is available on Agentic/Enterprise but is confined to the customer's deployment and never feeds Tabnine's shared models.","billing_note":"Prices are per user/month on annual subscription (no monthly-billing rate published; \"annual subscription\" stated for both tiers). No free/Basic tier — Tabnine retired the free tier in 2024. Annual discount exists but rate not disclosed (contact sales). No published minimum-seat count. When using Tabnine-provided LLM access, token-consumption fees apply on top (actual LLM provider price + ~5% handling fee); unlimited usage when bringing your own on-prem/cloud LLM endpoint. Enterprise tier is custom-quoted.","coding_quality":3,"confidence":"high","link_status":"live","swe_bench":0},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":false,"soc2":true},"name":"Google AI (Pro / Ultra / Ultra Max)","link":"https://gemini.google/subscriptions/","sources":["https://gemini.google/subscriptions/","https://9to5google.com/2026/04/11/google-ai-pro-ultra-features/","https://vibecoding.app/blog/google-antigravity-pricing-2026","https://antigravity.google/terms","https://discuss.ai.google.dev/t/how-can-i-completely-opt-out-of-the-use-of-my-data-for-model-training/168429","https://www.nxcode.io/resources/news/gemini-3-1-pro-complete-guide-benchmarks-pricing-api-2026","https://www.eesel.ai/blog/google-gemini-3-pricing","https://9to5google.com/2026/05/05/google-ai-ultra-lite-gemini-usage-limits/"],"category":"model_provider","endpoint":"native","region":"US","slug":"google-ai","model":"Gemini 3.1 Pro (+ Jules, Antigravity)","note":"Consumer tiers verified on live gemini.google/subscriptions: Pro $19.99, Ultra 5x $99.99, Ultra 20x ('Ultra Max') $199.99, plus a Plus tier at $7.99 (seed omitted). Top tier was cut from $249.99 at I/O 2026. Coding stack = Gemini 3.1 Pro + Jules (async agent) + Antigravity (agentic IDE). Gemini 3.1 Pro = 80.6% SWE-bench Verified (seed's 76.2 is stale). KEY GOVERNANCE FLAG for a CITIC-group firm: the seed's 'no-train' is WRONG for consumer plans — Antigravity/Jules/Gemini default to training on Interactions (incl. source code) with human review; opt-out exists but is OFF by default, and true zero-data-retention requires Google Workspace / Vertex AI (Workspace sign-in for Antigravity not yet GA). Enterprise admin/SSO/SAML/SCIM, SOC 2 / ISO 27001 available via Google Workspace + Vertex AI; no self-host/air-gap (cloud only). Google-served, reachable from Hong Kong without VPN or US tenant. For deployment, route through Workspace/Vertex to get ZDR + no-train. ⚠ verify Ultra-Lite tier.","tiers":[{"label":"Free","unit":"seat","key":"free","limit":"Baseline Gemini access; lowest usage limits; entry Jules/Antigravity quota","monthly":0},{"label":"Google AI Plus","unit":"seat","key":"plus","limit":"~2x Free usage limits","monthly":7.99},{"label":"Google AI Pro","unit":"seat","key":"pro","limit":"~4x Free; Gemini 3.1 Pro, Jules, Antigravity, AI Studio vibe-coding, Android Studio","monthly":19.99},{"label":"Google AI Ultra (5x)","unit":"seat","key":"ultra","limit":"~5x Pro quotas; Deep Think, Veo, ~25k monthly AI credits, higher Jules/Antigravity agent limits","monthly":99.99},{"label":"Google AI Ultra (20x / 'Ultra Max')","unit":"seat","key":"ultra-max","limit":"~20x Pro quotas; highest Jules/Antigravity rate limits, first access to new models","monthly":199.99}],"trains_on_code":"inputs","hk_ok":true,"hosting":"Google One subscription","trains_scope":"Antigravity Additional ToS: by default Google uses 'Interactions' (incl. source code, prompts, model responses fed to the agent) to improve products and ML, with human review — i.e. trains on your code unless you opt out. Consumer Gemini/Jules same regime (opt out via Gemini Apps Activity). True no-train/ZDR only via Google Workspace or Vertex AI / GCP; Workspace sign-in for Antigravity not yet fully supported. Code training ON by default for consumer tiers, opt-out (not opt-in) available; enterprise Workspace/Vertex never trained.","billing_note":"Prices are USD/month per individual seat (Pro $19.99, Ultra 5x $99.99, Ultra 20x $199.99; top tier cut from $249.99 at I/O 2026). Page geo-serves local currency/150+ countries (e.g. SGD 28.99 Pro, SGD 139.99 / 289.98 Ultra). No public annual-discount or min-seat terms on the consumer page; business pricing is via Google Workspace / Gemini Enterprise / Vertex AI sales. A rumored 'AI Ultra Lite' tier was reported but not yet live. ⚠ verify Ultra-Lite","coding_quality":5,"confidence":"high","link_status":"live","swe_bench":80.6},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":false,"soc2":true},"name":"xAI SuperGrok","link":"https://x.ai/grok","sources":["https://felloai.com/grok-pricing/","https://aitoolanalysis.com/supergrok-subscription-price-2026/","https://www.techradar.com/pro/elon-musk-goes-all-in-on-smbs-with-new-grok-enterprise-and-business-offerings","https://www.blockchain-council.org/ai/grok-business-enterprise-plans/","https://coursiv.io/blog/how-to-use-grok-for-business-2026","https://trustscan.dev/blog/how-to-opt-out-grok-xai-data-training-2026","https://x.ai/news/grok-code-fast-1","https://news.ycombinator.com/item?id=45064606","https://costbench.com/software/ai-chatbots/grok/"],"category":"model_provider","endpoint":"native","region":"US","slug":"xai-supergrok","model":"Grok 4.x (Grok 4.3 / Grok 4 Heavy)","note":"RE-VERIFIED on 2026-dated sources. Consumer ladder confirmed: Free $0 / SuperGrok Lite $10 (cheapest paid, late Mar 2026) / SuperGrok $30 ($300/yr) / SuperGrok Heavy $300. Org tiers confirmed (TechRadar, Blockchain Council, Coursiv): Grok Business $30/user/mo self-serve and Grok Enterprise custom — SSO/SCIM, audit logs, customer-managed keys, dedicated data plane, and an explicit 'no training on your data' guarantee. For CLSA, Business/Enterprise is the compliant path: code NOT used for training there (consumer SuperGrok trains chats by default with opt-out + EU GDPR objection). SOC 2 ⚠ verify — enterprise security controls confirmed but explicit SOC 2 cert not found on a 2026 source. Coding: grok-code-fast-1 70.8% SWE-bench Verified on xAI's OWN internal harness (HN + x.ai) — independent harness scores vary ⚠ verify. US-hosted, globally reachable, no US-tenant lock — HK-usable. No self-host/on-prem/air-gap. Official x.ai/grok is LIVE but returned HTTP 403 (Cloudflare) to the fetcher — link_status=unreadable; all figures corroborated across multiple 2026 third-party sources; verify exact prices/ctx/SWE-bench on x.ai directly.","tiers":[{"label":"Free","unit":"seat","key":"free","limit":"Grok 3-class access, low rate limits; entry tier","monthly":0},{"label":"SuperGrok Lite","unit":"seat","key":"supergrok-lite","limit":"Cheapest paid tier (launched late Mar 2026); 1 AI agent, 2x longer convos, basic image/video (480p, ~6s clips), modest limits","monthly":10},{"label":"SuperGrok","unit":"seat","key":"supergrok","limit":"Full Grok 4 / 4.3 (staged), DeepSearch, Big Brain, Voice; 128K ctx; $300/yr (~17% off)","monthly":30},{"label":"SuperGrok Heavy","unit":"seat","key":"supergrok-heavy","limit":"Grok 4 Heavy multi-agent + full Grok 4.3, max rate limits, ~1000 images/100 video clips, API access ⚠ verify ctx/quotas","monthly":300},{"label":"Grok Business","unit":"seat","key":"business","limit":"Org tier: team mgmt, central billing, no training on your data, connectors; self-serve $30/user/mo","monthly":30},{"label":"Grok Enterprise","unit":"seat","key":"enterprise","limit":"Custom pricing: SSO/SCIM, audit logs, customer-managed keys, dedicated data plane, advanced models, no training on your data","monthly":0}],"trains_on_code":"no","hk_ok":true,"hosting":"x.ai subscription","trains_scope":"Consumer (Free/SuperGrok) chats train xAI models by default, but a settings opt-out exists and is honored (X > Settings > Privacy > Grok); EU users can object under GDPR Art. 21. Grok Business & Enterprise explicitly guarantee no training on your data; Enterprise adds data isolation / customer-managed encryption keys. For a firm on Business/Enterprise, source code is not used for training (no = prompt/code never trained on those tiers).","billing_note":"SuperGrok $30/mo or $300/yr (~17% annual discount, saves ~$60). SuperGrok Heavy $300/mo, no annual option published. SuperGrok Lite $10/mo (added late Mar 2026, cheapest paid tier). X Premium ($8) and X Premium+ ($40) are X social bundles that also include Grok. Grok Business $30/user/mo self-serve; Grok Enterprise custom/sales-quoted. ⚠ verify exact seat minimums — none published. ⚠ verify Heavy context window (128K vs 256K) on x.ai.","coding_quality":4,"confidence":"medium","link_status":"unreadable","swe_bench":70.8},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":true,"soc2":false},"name":"Mistral Le Chat Pro","link":"https://mistral.ai/pricing","sources":["https://mistral.ai/pricing","https://help.mistral.ai/en/articles/455207-can-i-opt-out-of-my-input-or-output-data-being-used-for-training","https://mlq.ai/news/mistral-rebrands-le-chat-as-vibe-launches-autonomous-work-agent-for-enterprise/","https://the-decoder.com/mistral-rebrands-lechat-as-vibe-betting-its-chatbots-future-is-as-a-full-blown-work-agent/","https://www.cloudzero.com/blog/mistral-api-pricing/","https://costbench.com/software/ai-chatbots/mistral/","https://felloai.com/mistral-ai-pricing/","https://www.marktechpost.com/2026/05/02/mistral-ai-launches-remote-agents-in-vibe-and-mistral-medium-3-5-with-77-6-swe-bench-verified-score/","https://mistral.ai/news/devstral-2-vibe-cli/","https://help.mistral.ai/en/articles/393497-do-you-use-my-connectors-data-for-training"],"category":"model_provider","endpoint":"native","region":"EU","slug":"mistral-le-chat","model":"Codestral (in Le Chat)","note":"Le Chat was rebranded to \"Vibe\" (effective ~June 2026); pricing page is live at mistral.ai/pricing. Pro unchanged at $14.99/mo and now bundles Code Mode / Vibe CLI (Mistral Medium 3.5, 77.6% SWE-bench Verified; Devstral 2 lineage) plus autonomous Work Mode. EU-hosted (Microsoft/Google-independent), globally reachable from HK with no US tenant — good fit for a CITIC-group firm. No-train is opt-out on Pro (toggle), default on Team/Enterprise. Enterprise adds SAML SSO + self-host/on-prem/private cloud + EU residency + audit logs. SOC 2 not stated on pricing page (Mistral generally holds ISO 27001 / SOC 2 ⚠ verify). Annual Team discount from third-party sources ⚠ verify.","tiers":[{"label":"Free","unit":"seat","key":"free","limit":"$0; limited messages, web searches, coding sessions & image gen under fair-use caps (~25 msgs/day)","monthly":0},{"label":"Pro","unit":"seat","key":"pro","limit":"More messages/web search; all-day coding in CLI, IDE & web (Code Mode); deep research; image gen; No Telemetry toggle","monthly":14.99},{"label":"Team","unit":"seat","key":"team","limit":"$24.99/user/mo (~$19.99 annual ⚠ verify); up to 30GB storage/user, shared libraries, admin console, domain verification, data export; opted out of training by default","monthly":24.99},{"label":"Enterprise","unit":"seat","key":"enterprise","limit":"Custom pricing; custom models/agents/workflows, audit logs, SAML SSO, white-label, EU data residency, self-host/on-prem/private cloud; opted out of training by default","monthly":0}],"trains_on_code":"no","hk_ok":true,"hosting":"mistral.ai subscription","trains_scope":"Free/Pro are opted INTO training on inputs/outputs (incl. code) by default but offer a No Telemetry / opt-out toggle; Team & Enterprise are opted OUT of training by default. Connector-accessed data is never used for training on any plan. So code training is fully avoidable.","billing_note":"Pro $14.99/seat/mo; Team $24.99/seat/mo monthly, ~$19.99/seat/mo billed annually (third-party reported ⚠ verify); no published minimum-seat count; prices ex-tax; EU listings show same figures in € (€14.99 / €24.99). Enterprise custom (median ~$24K/yr deal-flow ⚠ verify).","coding_quality":4,"confidence":"high","link_status":"live","swe_bench":77.6},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":false,"soc2":true},"name":"Amazon Q Developer","link":"https://aws.amazon.com/q/developer/pricing/","sources":["https://aws.amazon.com/q/developer/pricing/","https://aws.amazon.com/q/developer/faqs/","https://aws.amazon.com/blogs/devops/amazon-q-developer-end-of-support-announcement/","https://aws.amazon.com/q/developer/features/","https://kiro.dev/pricing/"],"category":"model_provider","endpoint":"native","region":"global","slug":"amazon-q-developer","model":"Amazon Q (Bedrock-backed)","note":"PRODUCT BEING SUNSET — do not adopt for new rollout. New signups (Free + Pro) blocked 15 May 2026; full end-of-support 30 Apr 2027. AWS is migrating users to Kiro (the successor IDE; latest Opus 4.7 exclusive to Kiro, Opus 4.6 dropped from Q Pro 29 May 2026). Pricing (still live): Free $0 (50 agentic requests/mo); Pro $19/user/mo with admin dashboard + IP indemnity. Training: Pro does NOT train on customer code/content and excludes it from service improvement; Free Tier MAY use code/content unless opted out — so use Pro only. SSO via AWS IAM Identity Center; SOC 2/ISO 27001 via AWS. No on-prem/air-gap (runs in AWS cloud, Bedrock-backed). Reachable from Hong Kong via standard AWS — no US tenant/VPN needed (an AWS-account-level service). SWE-bench 66% ⚠ verify (Apr-2025 self-reported figure, now dated). For a CITIC-group firm, recommend evaluating Kiro instead given the EOL timeline.","tiers":[{"label":"Free Tier","unit":"seat","key":"free","limit":"50 agentic requests/mo; latest Claude models in IDE & CLI; data may be used to train FMs unless opted out","monthly":0},{"label":"Pro","unit":"seat","key":"pro","limit":"Higher agentic request limit; admin dashboard, IP indemnity, no-train; 4,000 LOC/mo transform pooled at account, $0.003/LOC overage","monthly":19}],"trains_on_code":"no","hk_ok":true,"hosting":"AWS","trains_scope":"Pro Tier: content/code NOT used to train underlying FMs and not used for service improvement. Free Tier MAY use content (incl. code) for FM training/service improvement unless explicitly opted out (AWS Organizations AI opt-out policy or IDE setting). AWS-wide: customer content never used to train base models.","billing_note":"$19/user/mo, billed monthly per user, pro-rated for partial months but cancellation mid-month incurs full month. No published annual discount or minimum seats. Code-transformation overage $0.003/LOC beyond 4,000 LOC/mo pooled at AWS payer-account level. SUNSET: new signups (Free + Pro) blocked 15 May 2026; full end-of-support 30 Apr 2027; existing Pro subs can still add users until then. AWS steering customers to Kiro (Pro $20/mo, latest Opus 4.7 exclusive to Kiro).","coding_quality":3.5,"confidence":"high","link_status":"live","swe_bench":66},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":false,"soc2":true},"name":"OpenAI ChatGPT","link":"https://chatgpt.com/pricing","sources":["https://techcrunch.com/2026/04/09/chatgpt-pro-plan-100-month-codex/","https://thenextweb.com/news/openais-new-100-chatgpt-pro-plan-targets-claude-max-with-five-times-the-codex-access","https://venturebeat.com/orchestration/openai-introduces-chatgpt-pro-usd100-tier-with-5x-usage-limits-for-codex","https://www.cloudzero.com/blog/how-much-does-chatgpt-cost/","https://www.finout.io/blog/openai-pricing-in-2026","https://techjacksolutions.com/ai-tools/chatgpt/chatgpt-pricing/","https://openai.com/enterprise-privacy/","https://help.openai.com/en/articles/7947663-chatgpt-supported-countries","https://www.scmp.com/tech/policy/article/3267971/tech-war-openai-further-block-access-mainland-china-hong-kong-based-developers","https://www.ptsconsulting.com.hk/blog/openai-ban"],"category":"model_provider","endpoint":"openai","region":"US","slug":"openai-chatgpt","model":"GPT-5 (+ Codex)","note":"Re-verified June 2026. Canonical pricing at chatgpt.com/pricing — official page returns HTTP 403 to fetcher (bot-blocked / unreadable), so figures corroborated from multiple 2026 secondary sources; treat live figures as ⚠ verify. Plan count CONFIRMED at 7: Free $0 / Go $8 / Plus $20 / Pro (Codex) $100 / Pro (Max) $200 / Business $25 mo ($20 annual, min 2) / Enterprise custom. Cheapest paid = Go $8 ⚠ verify; top tier = Pro (Max) $200. $100 Pro (Codex) tier confirmed launched 9 Apr 2026 (5x Plus Codex; 10x promo to 31 May 2026, now likely lapsed ⚠ verify). trains_on_code=no: consumer training is prompt/conversation-level opt-out, not repo/source training; Business/Enterprise/API no-train by default. SOC 2 Type II + ISO 27001; SSO/SAML on Business, SCIM on Enterprise. No self-host/air-gap. HK BLOCKER confirmed: OpenAI geofences Hong Kong — no +852 signup, HK IPs blocked; not usable from HK without non-HK tenant/VPN. Compliant route for a CITIC-group firm is Azure OpenAI / Microsoft Copilot (models hosted outside HK). hk_ok=false. SWE-bench: no official ChatGPT-product figure (GPT-5.x model scores exist separately) → 0.","tiers":[{"label":"Free","unit":"seat","key":"free","limit":"GPT-5.5 with tight caps; basic search, limited file/image, no priority access","monthly":0},{"label":"Go","unit":"seat","key":"go","limit":"Higher limits than Free; consumer tier (global rollout since Jan 2026, 160+ countries)","monthly":8},{"label":"Plus","unit":"seat","key":"plus","limit":"Expanded GPT-5.5 / Thinking limits, standard Codex, ~10 deep research/mo, image/Sora basic, Agent mode","monthly":20},{"label":"Pro (Codex)","unit":"seat","key":"pro-codex","limit":"~5x Plus quotas + 5x Plus Codex usage (10x promo to 31 May 2026); all Pro models/features","monthly":100},{"label":"Pro (Max)","unit":"seat","key":"pro-max","limit":"Max limits: ~20x Plus, near-unlimited GPT-5.5/Pro, priority Codex, 250 deep-research/mo, Sora, ~1M context","monthly":200},{"label":"Business (formerly Team)","unit":"seat","key":"business","limit":"$25/seat monthly, $20 annual; min 2 seats; admin console, no-train default, SAML SSO, SOC 2","monthly":25},{"label":"Enterprise","unit":"seat","key":"enterprise","limit":"Custom pricing (~150+ seats); SCIM, data residency, CMEK, advanced admin/audit, no-train default","monthly":0}],"trains_on_code":"no","hk_ok":false,"hosting":"OpenAI","trains_scope":"Consumer Free/Plus/Pro: conversations (incl. pasted code) may be used to train models unless the user opts out / disables chat history. Business, Enterprise, Edu and the API do NOT train on workspace inputs/outputs by default; Enterprise supports custom retention/ZDR. No training on source repos at org tiers.","billing_note":"Business ~20% cheaper annually ($20 vs $25/seat/mo), min 2 seats. Pro (Codex) $100 launched 9 Apr 2026: 5x Plus Codex usage (10x promo through 31 May 2026). Go $8 consumer tier. Enterprise custom, typically ~150+ seats. USD.","coding_quality":4.5,"confidence":"medium","link_status":"unreadable","swe_bench":0},{"flags":{"enterprise":true,"sso":true,"air_gap":false,"self_host":false,"soc2":true},"name":"Anthropic Claude","link":"https://claude.com/pricing","sources":["https://claude.com/pricing","https://www.anthropic.com/legal/commercial-terms","https://docs.anthropic.com/en/docs/claude-code/data-usage","https://www.lexology.com/library/detail.aspx?g=619e126a-e78e-475d-97d9-d6067f1505b6","https://www.vellum.ai/blog/claude-opus-4-8-benchmarks-explained","https://runbear.io/posts/claude-enterprise-pricing-mid-market","https://tygartmedia.com/claude-team-pricing-2026-standard-premium-seats/"],"category":"model_provider","endpoint":"anthropic","region":"US","slug":"anthropic-claude","model":"Claude (Opus/Sonnet, + Claude Code)","note":"Official pricing now at claude.com/pricing (moved from anthropic.com), LIVE. Plans: Free $0, Pro $20 ($17 annual), Max 5x $100, Max 20x $200, Team $25/seat ($20 annual; premium $125/$100), Enterprise self-serve $20/seat + API usage or custom. CRITICAL CORRECTION vs prior 'no-train': consumer Pro/Max DO train on chats and CODE sessions by default since Sept 2025 (5-yr retention) unless opted out — for a financial firm, mandate Team/Enterprise (no-train by default) and avoid consumer seats for repo work. Flagship Opus 4.8 (May 2026) scores 88.6% SWE-bench Verified (seed 77.2 was stale). Claude Code now ships on Pro+ and is reachable from HK without VPN (prior 'blocked in HK' note outdated). SOC 2 Type II / ISO 27001 attested; SSO/SCIM on Team/Enterprise. No self-host/air-gap (cloud only — relevant for CITIC/mainland data-residency review; consider Bedrock for VPC isolation). Enterprise pricing and the $1,000/seat promo ⚠ verify.","tiers":[{"label":"Free","unit":"seat","key":"free","limit":"Web/mobile chat, limited daily usage, web search, memory, connectors","monthly":0},{"label":"Pro","unit":"seat","key":"pro","limit":"Higher limits; includes Claude Code, Cowork, Design. $17/mo if billed annually","monthly":20},{"label":"Max 5x","unit":"seat","key":"max-5x","limit":"~5x Pro usage; priority access. Monthly billing only","monthly":100},{"label":"Max 20x","unit":"seat","key":"max-20x","limit":"~20x Pro usage; priority access. Monthly billing only","monthly":200},{"label":"Team","unit":"seat","key":"team","limit":"Standard seat $25/mo ($20 annual); Premium seat $125/mo ($100 annual). 5-150 users. SSO, central billing, no-train by default","monthly":25},{"label":"Enterprise","unit":"seat","key":"enterprise","limit":"Self-serve $20/seat + API-rate usage; sales-assisted custom. SCIM, audit logs, HIPAA-ready, data residency","monthly":20}],"trains_on_code":"inputs","hk_ok":true,"hosting":"Anthropic","trains_scope":"Consumer Free/Pro/Max (incl. Claude Code sessions) train on chats AND coding sessions BY DEFAULT since Aug-Sep 2025, retained up to 5 years unless user opts out (opt-out keeps 30-day retention; safety-flagged convos exempt from opt-out). Team/Enterprise/API/commercial: NO training on customer content by default. For an enterprise buyer use Team/Enterprise tier to get no-train.","billing_note":"Pro $17/mo billed annually (vs $20 monthly). Max 5x/20x monthly-only. Team standard seat $20 annual / $25 monthly; premium seat $100 annual / $125 monthly; 5-150 seats. Enterprise self-serve $20/seat + usage at API rates. June 2026 promo: $1,000 Claude Code + Cowork credits per Enterprise seat activated by Jul 2, 2026. ⚠ verify promo.","coding_quality":5,"confidence":"high","link_status":"live","swe_bench":88.6}],"presets":[{"label":"Recommended","key":"recommended","blurb":"GLM + Kimi, tier-sized by seniority — seniors add Kimi Allegro + Trae Enterprise for agent-heavy work","roles":{"senior":{"copilot":"enterprise","subs":[{"slug":"zai-glm-coding-plan","tier":"max"},{"slug":"moonshot-kimi-for-coding","tier":"allegro"},{"slug":"trae","tier":"enterprise"}]},"mid":{"copilot":"enterprise","subs":[{"slug":"zai-glm-coding-plan","tier":"pro"},{"slug":"moonshot-kimi-for-coding","tier":"allegretto"}]},"junior":{"copilot":"enterprise","subs":[{"slug":"zai-glm-coding-plan","tier":"lite"},{"slug":"moonshot-kimi-for-coding","tier":"allegretto"}]}}},{"label":"Best value","key":"best_value","blurb":"MiMo-led — cheapest strong coder per dollar","roles":{"senior":{"copilot":"enterprise","subs":[{"slug":"xiaomi-mimo-token-plan","tier":"pro"},{"slug":"zai-glm-coding-plan","tier":"pro"}]},"mid":{"copilot":"enterprise","subs":[{"slug":"xiaomi-mimo-token-plan","tier":"standard"},{"slug":"zai-glm-coding-plan","tier":"lite"}]},"junior":{"copilot":"enterprise","subs":[{"slug":"xiaomi-mimo-token-plan","tier":"standard"}]}}},{"label":"ByteDance · enterprise","key":"byteplus_enterprise","blurb":"Single-vendor BytePlus stack — Doubao coder (SWE 76.5) + Trae IDE; HK-ok, no code-training, enterprise","roles":{"senior":{"copilot":"enterprise","subs":[{"slug":"bytedance-doubao-byteplus","tier":"pro"},{"slug":"trae","tier":"pro"}]},"mid":{"copilot":"enterprise","subs":[{"slug":"bytedance-doubao-byteplus","tier":"lite"},{"slug":"trae","tier":"lite"}]},"junior":{"copilot":"enterprise","subs":[{"slug":"bytedance-doubao-byteplus","tier":"lite"}]}}},{"label":"Privacy · on-shore Qwen","key":"privacy_onshore","blurb":"Unlimited no-train Qwen, self-hosted on CITIC GPUs — nothing leaves the building","roles":{"senior":{"copilot":"enterprise","subs":[{"slug":"alibaba-qwen-coding-plan","tier":"onshore"}]},"mid":{"copilot":"enterprise","subs":[{"slug":"alibaba-qwen-coding-plan","tier":"onshore"}]},"junior":{"copilot":"enterprise","subs":[{"slug":"alibaba-qwen-coding-plan","tier":"onshore"}]}}},{"label":"Lingma near-zero","key":"lingma_zero","blurb":"Alibaba Lingma — Qwen-powered, no-train; free individual tier or $32 dedicated seat","roles":{"senior":{"copilot":"enterprise","subs":[{"slug":"alibaba-lingma","tier":"enterprise"}]},"mid":{"copilot":"enterprise","subs":[{"slug":"alibaba-lingma","tier":"individual"}]},"junior":{"copilot":"enterprise","subs":[{"slug":"alibaba-lingma","tier":"individual"}]}}},{"label":"Frontier-lite","key":"frontier","blurb":"Top tiers everywhere — GLM Max + Kimi Allegro for the heaviest agentic work","roles":{"senior":{"copilot":"enterprise","subs":[{"slug":"zai-glm-coding-plan","tier":"max"},{"slug":"moonshot-kimi-for-coding","tier":"allegro"}]},"mid":{"copilot":"enterprise","subs":[{"slug":"zai-glm-coding-plan","tier":"max"},{"slug":"moonshot-kimi-for-coding","tier":"allegretto"}]},"junior":{"copilot":"enterprise","subs":[{"slug":"zai-glm-coding-plan","tier":"pro"}]}}},{"label":"Lean","key":"lean","blurb":"Smallest SaaS-only footprint — one plan per dev, no self-host to run","roles":{"senior":{"copilot":"enterprise","subs":[{"slug":"zai-glm-coding-plan","tier":"pro"}]},"mid":{"copilot":"enterprise","subs":[{"slug":"xiaomi-mimo-token-plan","tier":"standard"}]},"junior":{"copilot":"enterprise","subs":[]}}}],"categories":[{"label":"Coding plans","key":"coding_plan","blurb":"Flat-rate open-weight / Chinese workhorses — the cheap daily drivers","icon":"hero-bolt"},{"label":"All-in-one IDEs & agents","key":"ide_agent","blurb":"Full editors with built-in models","icon":"hero-window"},{"label":"Terminal & autonomous agents","key":"terminal_agent","blurb":"BYOK harnesses and async cloud agents","icon":"hero-command-line"},{"label":"Model-provider subscriptions","key":"model_provider","blurb":"Frontier chat / coding subscriptions","icon":"hero-sparkles"},{"label":"Proxies & gateways","key":"proxy","blurb":"One key to many models","icon":"hero-arrows-right-left"}],"envelope":{"core":300,"proxy":50,"ceiling":389,"copilot":39},"proxy_models":[{"input":0.12,"name":"Gemma 4 (31B)","output":0.35,"context":"256K (262,144)","sources":["https://openrouter.ai/google/gemma-4-31b-it","https://designforonline.com/ai-models/google-gemma-4-31b/","https://pricepertoken.com/pricing-page/model/google-gemma-4-31b-it","https://artificialanalysis.ai/models/gemma-4-31b/providers","https://openrouter.ai/google/gemma-4-26b-a4b-it","https://en.wikipedia.org/wiki/Gemma_(language_model)"],"slug":"gemma-4","provider":"Google (open-weight; PAYG via third-party hosts)","note":"Held in/out rate ($0.12/$0.35 per 1M, non-cached/non-batch) CONFIRMED across OpenRouter, pricepertoken, designforonline. CONTEXT CORRECTION: window is 256K (262,144 tokens), not the held 128K. Apache-2.0 open weights confirmed. ARCHITECTURE NOTE: the 31B is a ~30.7B DENSE model, not MoE; the cheaper 26B-A4B ($0.06/$0.33, 256K) is a separate MoE variant (25.2B total / 3.8B active) — both confirmed. Released 2026-04-02. Caveat: Gemma is open-weight, so there is NO single official Google per-token list price; $0.12/$0.35 is the prevailing third-party/aggregator rate a LiteLLM proxy would route to (e.g. DeepInfra ~$0.13/$0.38). Prompt-caching can cut effective input cost 60-80%; no published batch tier. hk_ok=true: open-weight model on self-hosted LiteLLM has no geo API restriction.","hk_ok":true,"confidence":"high"},{"input":0.14,"name":"DeepSeek V4 Flash","output":0.28,"context":"1M","sources":["https://api-docs.deepseek.com/quick_start/pricing","https://openrouter.ai/deepseek/deepseek-v4-flash","https://devtk.ai/en/models/deepseek-v4-flash/","https://skywork.ai/skypage/en/deepseek-data-privacy-security-guide/2047585299882700800"],"slug":"deepseek-v4-flash","provider":"DeepSeek","note":"Confirmed on DeepSeek official pricing docs: $0.14/1M input (cache-miss), $0.28/1M output, 1M ctx (max output 384K). Cache-hit input is $0.0028/1M (98% off) auto-enabled, no batch tier listed. Held values correct. Trains-on-inputs: standard DeepSeek API stores data in China and may use it for training; self-hosted LiteLLM proxy still forwards prompts to DeepSeek's endpoint, so this data-residency/training caveat persists regardless of proxy/HK egress. hk_ok=true: no HK geo-block on the API and a self-hosted proxy avoids any client-side restriction.","hk_ok":true,"confidence":"high"},{"input":0.14,"name":"MiMo-V2.5","output":0.28,"context":"1M","sources":["https://x.com/XiaomiMiMo/status/2059314052892099070","https://gigazine.net/gsc_news/en/20260527-xiaomi-mimo-v2-5-price-down/","https://openrouter.ai/xiaomi/mimo-v2.5","https://developer.puter.com/ai/xiaomi/mimo-v2.5/","https://pricepertoken.com/pricing-page/model/xiaomi-mimo-v2.5","https://www.aipricing.guru/news/xiaomi-mimo-v2-5-api-price-cut-may-2026/","https://apidog.com/blog/xiaomi-mimo-v2-5-api-cost/","https://mimo.mi.com/docs/pricing"],"slug":"mimo-v2.5","provider":"Xiaomi (MiMo platform; via self-hosted LiteLLM proxy)","note":"CORRECTED from held values. Current PAYG list rate is $0.14 in / $0.28 out per 1M (non-cached, non-batch), UNIFIED across all context lengths after Xiaomi's permanent price cut effective 2026-05-27 (May 26, 6PM PDT). Held values ($0.40/$2.00, 256K) were the stale pre-cut MiMo-v2 long-context-band rates. Context window is 1M (not 256K). Cached-input (prompt cache hit) = $0.0028 per 1M (50x cheaper); some aggregators quote a higher cache rate (~$0.20/M) — official Xiaomi announcement states $0.0028. No batch discount published. Pro variant (MiMo-V2.5-Pro) is ~$0.435 in / $0.87 out per 1M — NOT the held ~$1.00/$3.00. The HK self-hosted LiteLLM proxy does not change list rates: LiteLLM passes through upstream Xiaomi token pricing; API is publicly reachable from Hong Kong, billed in USD, so hk_ok=true. Official docs page (platform.xiaomimimo.com -> mimo.mi.com) is JS-rendered and returned no fetchable pricing text; figures confirmed via official Xiaomi MiMo announcement (X/GIGAZINE/HN) + OpenRouter + Puter + pricepertoken + aipricing.guru.","hk_ok":true,"confidence":"high"},{"input":0.3,"name":"Doubao-Seed-2.0-Code","output":1.2,"context":"256K","sources":["https://tokenmix.ai/blog/doubao-seed-2-0-code-review-2026","https://howaiworks.ai/models/doubao-seed-code","https://artificialanalysis.ai/models/doubao-seed-code","https://zenmux.ai/volcengine/doubao-seed-code","https://crazyrouter.com/en/blog/doubao-seed-code-api-guide","https://docs.byteplus.com/en/docs/ModelArk/1544106"],"slug":"doubao-seed-2.0-code","provider":"ByteDance / BytePlus (ModelArk / Volcano Engine)","note":"Coding-specialist model, 256K ctx (multi-source confirmed), 64K max output, Anthropic-API-compatible so works via self-hosted LiteLLM from HK. Held $0.30/$1.20 per 1M matches the TokenMix review but that source self-flags as 'subject to confirmation', and NO official BytePlus/Volcano Engine USD list page was scrapable to confirm. Aggregators diverge sharply, esp. on output: HowAIWorks $0.18/$0.71 (cached input $0.03; stated as conversion of ¥1.2/¥4.8), Crazyrouter reseller $0.30/$2.00, ZenMux blended ranges input $0.17-0.39 / output $1.12-2.25. ByteDance/Volcano prices are TIERED by input length (CN 0-32K tier ~¥1.2 in / ¥8.0 out per 1M) and offer large cache-hit discounts; batch/promo new-user trials also exist. Input ~$0.30 is plausible as BytePlus intl list rate; output $1.20 is the least certain figure (real list rate may be higher, ~$1.7-2.3). ⚠ verify against the BytePlus ModelArk console pricing page before relying on these.","hk_ok":true,"confidence":"low"},{"input":0.435,"name":"DeepSeek V4 Pro","output":0.87,"context":"1M","sources":["https://devtk.ai/en/models/deepseek-v4/","https://api-docs.deepseek.com/quick_start/pricing","https://www.morphllm.com/deepseek-v4","https://codersera.com/blog/deepseek-v4-pro-review-benchmarks-pricing-2026/","https://pricepertoken.com/pricing-page/model/deepseek-deepseek-v4-pro"],"slug":"deepseek-v4-pro","provider":"DeepSeek","note":"Standard PAYG list rate: $0.435/1M input (cache-miss), $0.87/1M output. Held $0.44 input was a round-up — exact is $0.435. Cached (cache-hit) input: $0.003625/1M. This rate is the now-permanent \"1/4-of-original\" price (was a 75% promo that became official after 2026-05-31, now in effect); original reference was $1.74/$3.48 per 1M. No separate batch tier published. 1M context, 384K max output. SWE-bench Verified 80.6 confirmed. hk_ok: DeepSeek API has no HK geo-restriction and a self-hosted LiteLLM proxy adds none.","hk_ok":true,"confidence":"high"},{"input":0.5,"name":"Qwen3.6 Plus","output":3,"context":"1M","sources":["https://artificialanalysis.ai/models/qwen3-6-plus/providers","https://openrouter.ai/qwen/qwen3.6-plus","https://llm-stats.com/models/qwen3.6-plus","https://www.alibabacloud.com/help/en/model-studio/models","https://dev.to/purpledoubled/qwen36-scores-734-on-swe-bench-with-only-3b-active-parameters-heres-why-that-matters-2fmp"],"slug":"qwen3.6-plus","provider":"Alibaba (Qwen / Model Studio)","note":"CORRECTED. Standard non-cached/non-batch list rate is $0.50 in / $3.00 out per 1M (Artificial Analysis tracks Alibaba Cloud as the provider; corroborated by Together + llm-stats). The held $0.33/$1.95 is NOT the list rate — it is OpenRouter's promotional \"35% off\" price ($0.50x0.65=$0.325, $3.00x0.65=$1.95). Context window is 1M tokens (max output 65,536), NOT 256K — the 256K figure is an input-length pricing-tier threshold in Alibaba's docs, not the context size. SWE-bench Verified ~78.8 confirmed. Alibaba uses tiered input-length pricing (0-256K cheaper than 256K-1M); cached-input and batch discounts available on Model Studio. hk_ok: Model Studio international (Singapore) endpoint reachable from HK via self-hosted LiteLLM, no geo-block. Caveat: $0.50/$3.00 not yet visible on Alibaba's own (US-cached) Model Studio docs page, which still listed only Qwen3.5-Plus; release-date varies by aggregator (Mar 31 vs Apr 2 2026). ⚠ verify","hk_ok":true,"confidence":"medium"},{"input":0.6,"name":"MiniMax M3","output":2.4,"context":"1M","sources":["https://www.minimax.io/blog/minimax-m3","https://wavespeed.ai/blog/posts/minimax-m3-price/","https://llm-stats.com/models/minimax-m3","https://openrouter.ai/minimax/minimax-m3","https://devtk.ai/en/models/minimax-m3/"],"slug":"minimax-m3","provider":"MiniMax","note":"Standard PAYG list rate for input <=512K tokens: $0.60/$2.40 per 1M (in/out). 7-day launch promo = 50% off: $0.30/$1.20. Cached input ~$0.12/M list (~$0.06/M promo). Long-context tier (input >512K, up to 1M) bills the whole request at exactly 2x standard: $1.20/$4.80 list ($0.60/$2.40 promo); cached read $0.24/M list. Max output 512K tokens; thinking/non-thinking modes share pricing. HK: international platform minimax.io reached via self-hosted LiteLLM proxy, so HK origin is fine. Held value ($0.30/$1.20 promo, $0.60/$2.40 std, 1M ctx) confirmed correct.","hk_ok":true,"confidence":"high"},{"input":0.6,"name":"GLM-4.7","output":2.2,"context":"203K","sources":["https://docs.z.ai/guides/overview/pricing","https://openrouter.ai/z-ai/glm-4.7","https://llm-stats.com/models/glm-4.7"],"slug":"glm-4.7","provider":"Z.ai (Zhipu AI)","note":"Confirmed against Z.ai official pricing docs: standard non-cached PAYG list rate is $0.60 in / $2.20 out per 1M tokens. Cached input billed at ~$0.11/1M (cache storage currently \"limited-time free\"). Held values verified correct. OpenRouter lists a lower ~$0.40/$1.75 (OpenRouter's own discounted rate, not Z.ai's official list). 203K (~202.8K max input) context confirmed. Reachable from HK via self-hosted LiteLLM proxy against the z.ai international endpoint; no API geo-block.","hk_ok":true,"confidence":"high"},{"input":0.95,"name":"Kimi K2.7-Code","output":4,"context":"256K (262,144)","sources":["https://platform.kimi.ai/docs/pricing/chat-k27-code","https://platform.moonshot.ai/docs/pricing/chat","https://tokencost.app/blog/kimi-k2-7-code-pricing","https://openrouter.ai/moonshotai/kimi-k2.7-code","https://platform.kimi.ai/docs/agreement/userprivacy","https://platform.moonshot.ai/docs/agreement/modeluse"],"slug":"kimi-k2.7-code","provider":"Moonshot AI (Kimi OpenPlatform)","note":"Confirmed on official Moonshot/Kimi platform pricing page (platform.moonshot.ai redirects to platform.kimi.ai): standard non-cached list rate $0.95 in / $4.00 out per 1M; cache-hit input $0.19/1M (up from K2.6's $0.16); no separate batch rate published for this model. Context 262,144 tokens. All HELD values correct. Moonshot's Kimi OpenPlatform privacy policy/ToS allows using API content to \"develop and improve the services\" (i.e. trains on inputs) with no API opt-out. hk_ok: no geo-restriction blocks Hong Kong; a self-hosted LiteLLM proxy just forwards to Moonshot, so HK access is fine. Note data-residency/training-on-inputs concern for sensitive workloads. OpenRouter lists lower blended provider rates ($0.68/$3.41) — those are third-party/discounted, not the Moonshot list rate.","hk_ok":true,"confidence":"high"},{"input":1,"name":"Qwen3-Coder (Plus)","output":5,"context":"1M","sources":["https://www.alibabacloud.com/help/en/model-studio/models","https://www.alibabacloud.com/help/en/model-studio/qwen-coder","https://cloudprice.net/models/alibaba-qwen3-coder-plus","https://developer.puter.com/tutorials/qwen-api-pricing/"],"slug":"qwen3-coder","provider":"Alibaba Cloud Model Studio","note":"CORRECTION: our held $0.65/$3.25 + 256K are wrong. Alibaba's OFFICIAL Model Studio (Intl) uses TIERED billing by input size; base tier (input ≤32K) list rate = $1.00 in / $5.00 out per 1M. Higher tiers: 32K-128K $1.80/$9.00; 128K-256K $3.00/$15.00; 256K-1M $6.00/$60.00 (whole request bills at the tier set by its input size). Context window = 1M tokens (256K was a tier threshold, not the window); max output 65,536. The $0.65/$3.25 figure is a third-party reseller (OpenRouter) rate, NOT the official PAYG list rate a self-hosted LiteLLM proxy hits when calling Alibaba directly. Cached input: implicit cache 20% of standard, explicit cache 10%. Batch API ~$0.50/$2.50 (≈50% off, per aggregators). The \"480B base ~$0.22/$1.80\" refers to the separate open-weight Qwen3-Coder-480B-A35B model, not Plus. hk_ok: Model Studio International endpoint is publicly reachable from Hong Kong via API key; no in-region block.","hk_ok":true,"confidence":"high"},{"input":1.4,"name":"GLM-5.2","output":4.4,"context":"1M","sources":["https://docs.z.ai/guides/overview/pricing","https://openrouter.ai/z-ai/glm-5.2","https://cloudprice.net/models/zhipu-glm-5-2","https://www.requesty.ai/models/zai/glm-5.2","https://www.kucoin.com/news/flash/z-ai-launches-glm-5-2-with-1m-token-context-window-on-hugging-face"],"slug":"glm-5.2","provider":"Z.ai (Zhipu AI)","note":"Standard PAYG list rate confirmed on Z.ai's official pricing docs: $1.40/1M input, $4.40/1M output. Cached input $0.26/1M with limited-time free cache storage promo. No batch tier published. 1M-token context (max output 128K). Z.ai international platform serves Hong Kong; LiteLLM proxy compatible via OpenAI-style endpoint.","hk_ok":true,"confidence":"high"},{"input":2,"name":"Gemini 3.1 Pro","output":12,"context":"1M","sources":["https://ai.google.dev/gemini-api/docs/pricing","https://devtk.ai/en/models/gemini-3-1-pro/","https://www.metacto.com/blogs/the-true-cost-of-google-gemini-a-guide-to-api-pricing-and-integration","https://pricepertoken.com/pricing-page/model/google-gemini-3.1-pro-preview"],"slug":"gemini-3.1-pro","provider":"Google","note":"Standard non-cached, non-batch list rate for prompts ≤200K tokens. Tiered: prompts >200K cost $4.00 in / $18.00 out. Cached input $0.20 (≤200K) / $0.40 (>200K); cache storage $4.50 per 1M tokens/hour. Batch API 50% off: $1.00 in / $6.00 out (≤200K). Model is \"Gemini 3.1 Pro Preview\". hk_ok: Gemini paid API has no HK geo-block and a self-hosted LiteLLM proxy makes provider-side region moot.","hk_ok":true,"confidence":"high"},{"input":5,"name":"GPT-5.5 (via Azure)","output":30,"context":"1M","sources":["https://www.requesty.ai/models/azure/gpt-5.5-swedencentral","https://custom.typingmind.com/tools/estimate-llm-usage-costs/azure/gpt-5-5","https://the-decoder.com/openai-unveils-gpt-5-5-claims-a-new-class-of-intelligence-at-double-the-api-price/","https://azure.microsoft.com/en-us/pricing/details/azure-openai/"],"slug":"gpt-5.5-azure","provider":"Microsoft Azure OpenAI","note":"Standard PAYG list rate $5.00 in / $30.00 out per 1M confirmed across aggregators and OpenAI's GPT-5.5 announcement (2x prior-gen API price). Cached input ~$0.50/1M (90% off). Batch API standard 50% discount halves rates. Context window CORRECTED: held value 400K is wrong — Azure/OpenAI GPT-5.5 exposes ~1M+ tokens (≈922K input / 128K output); 400K is the Codex variant only. hk_ok: Azure OpenAI + self-hosted LiteLLM proxy reachable from Hong Kong (no provider geo-block). Held input/output rates verified correct. ⚠ verify — could not load Microsoft's official azure.microsoft.com pricing page (timeout/403); figures rest on third-party aggregators + OpenAI announcement, not the canonical Azure page.","hk_ok":true,"confidence":"medium"}],"roles":[{"label":"Senior","key":"senior","blurb":"power users · agent-heavy","default_count":20},{"label":"Mid-level","key":"mid","blurb":"steady daily driver","default_count":20},{"label":"Junior","key":"junior","blurb":"lighter · completion-led","default_count":10}]}