mirror of
https://github.com/Heretek-AI/Token-Max.git
synced 2026-10-12 02:03:18 +00:00
89 KiB
89 KiB
📑 SOURCE.md — Primary Evidence, Stated Usage Limits & Source Index
Purpose: This document is the comprehensive, audit-ready index of primary documentation sources, stated quota limits, and token derivation evidence for all 33 coding services and model providers tracked by Token-Max.
Audit Date: September 2026 | Scope: 33 Services · 414 Normalized Usage Limits · 107 Unit Tests | Enforced by:
npm run validate-data
Verification Standards & Metadata Schema
Every tier in Token-Max is classified with explicit provenance in estimateMeta:
-
Source Types:
official(direct vendor pricing/quota page),derived(mathematically inverted from official at-cost rates or credit ratios),research(published benchmarks, SWE-bench tracking, BSWEN 100M token study),community(forum telemetry, reverse engineering). -
Confidence Levels:
high(explicitly published token ceilings/tables),medium(official credit multipliers/at-cost API rates),low(opaque window/quota caps, research estimates). -
Stacking Policies:
allowed(explicitly permitted),silent(account sharing banned but multiple paid accounts unaddressed),prohibited(terms explicitly ban multiple accounts or proxying).
1. Coding IDEs & Autonomous Agents (14 Services)
Aider (aider)
- Primary Pricing & Docs: https://aider.chat
- Category:
coding-ide - Data Privacy & Training: Depends on API provider chosen
- IP Indemnity: False
- Stacking Policy:
unknown
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Free / BYOK | Free ($0) | toolCost: 100% Free Open Source Software rateLimits: Determined by your chosen API provider byok: Bring Your Own Key |
Claude Sonnet 5, GPT-6 Astra, DeepSeek V4.1 Flash, All OpenRouter models | Direct BYOK API billing (0% tool markup) Floor: 20M |
RESEARCH (low) |
Key Gotchas & Constraints:
- No hosted cloud backend — you pay the LLM API provider directly
- Large repo git-map indexing consumes significant context on startup
- Requires local Python environment and git repo
Amazon Q Developer (amazon-q)
- Primary Pricing & Docs: https://aws.amazon.com/q/developer/pricing/
- Category:
coding-ide - Data Privacy & Training: Opt-out available (Free); automatic opt-out (Pro)
- IP Indemnity: True
- Stacking Policy:
unknown
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Free Tier | Free ($0) | agenticRequests: 50 agentic requests/mo (chat + agentic coding) transformation: Java/.NET transformation 1,000 LOC/mo per user |
Latest Claude models via AWS | 50 agentic requests/mo (~1.1M tokens on the standard request basis) Floor: 1.05M |
DERIVED (high) Basis: standard 20K-in/1K-out agent request "Free tier: 50 agentic requests per month" |
| Pro | $19/mo | billing: $19/user/mo, activates on first agentic usage (pro-rata) requests: Higher agentic request limits (not numerically published) transformation: 4,000 LOC/mo pooled with $0.003/LOC overage |
Latest Claude models via AWS | Higher request pool (~21M floor / ~105M midpoint / ~210M ceiling scenarios) Floor: 21M |
Mid: 105M |
Key Gotchas & Constraints:
- Pro request cap is not numerically published - token estimate is research-derived
- AWS announced April 30, 2027 end-of-support for Amazon Q Developer IDE plugins; transitioning to Kiro as the successor agentic environment
- Subscription activates only on first agentic action or code completion
- Transformation overage billed at $0.003/LOC at payer-account level
- Data collection: opt-out available on Free; automatically opted out on Pro
- Deep-dive telemetry: see
docs/OSINT_USAGE_STATISTICS.md
Augment Code (augment-code)
- Primary Pricing & Docs: https://www.augmentcode.com/pricing
- Category:
coding-ide - Data Privacy & Training: No AI training on any plan
- IP Indemnity: False
- Stacking Policy:
silent— "Terms ban sharing credentials across users ("unique usernames and passwords cannot be shared or used by more than one individual Authorized User") but do not address multiple paid accounts."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Standard | $20/mo | billing: $20/mo flat per team (no per-seat charge), up to 50 seats usage: $20 of usage included/mo (LLM + Context Engine + compute) llmFee: Provider list price + flat 40% service fee on LLM usage topUps: PAYG top-ups valid 12 months |
Claude Opus 5, GPT-5.6 Sol, DeepSeek V4-Pro, Gemini 3.8 Flash, Cosmos | $20 pool with 40% fee (~7.1M tokens baseline; 1.49M Opus 5 to 205.05M Gemini Flash) Floor: 7.1M |
DERIVED (medium) Basis: $14.28 net credit ($20 / 1.4) across list-price models "flat 40% fee on LLM usage; $20 of usage included per month" |
| Business | $100/mo | billing: $100/mo flat per team, up to 50 seats usage: $100 of usage included/mo |
Claude Opus 5, GPT-5.6 Sol, DeepSeek V4-Pro, Gemini 3.8 Flash, Cosmos | $100 pool (~35.7M tokens baseline; 7.48M Opus 5 to 1025.66M Gemini Flash) Floor: 35.7M |
DERIVED (medium) Basis: $71.43 net credit ($100 / 1.4) across list-price models "$100 of usage included per month with the flat 40% LLM fee" |
| Enterprise | Enterprise | billing: Custom usage + top-ups; unlimited seats security: SSO/OIDC/SCIM |
Cosmos, API models | Custom usage contract Floor: 0M |
RESEARCH (low) |
Key Gotchas & Constraints:
- Flat 40% service fee on LLM usage means Standard ($20) provides $14.28 net compute, and Business ($100) provides $71.43 net compute; compute infrastructure is billed at cost (no fee)
- Multi-model catalog spans Claude Opus 5 (1.49M/7.48M), GPT-5.6 Sol (5.61M/28.04M), DeepSeek V4-Pro (43.23M/216.22M), Gemini 3.8 Flash (205.05M/1025.66M), and Cosmos
- Restructured from seat-based message plans to flat team pricing with dollar usage pools
- Top-up tokens valid 12 months
Claude Code (claude-code)
- Primary Pricing & Docs: https://claude.com/pricing
- Category:
coding-ide - Data Privacy & Training: Model training opt-out available on individual plans
- IP Indemnity: False
- Stacking Policy:
silent— "Consumer terms: "You may not share your Account login information... or make your Account available to anyone else"; multiple paid accounts are not addressed."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Pro | $20/mo ($200/yr) |
rollingCap: Shared 5-hour rolling session limit (>=5x Free tier) weeklyAllocation: Weekly caps on top of the 5-hour window overage: Opt-in usage credits billed at standard API rates, with optional monthly spend cap |
Claude Opus 5, Claude Sonnet 5, Claude Haiku 4.5 | Shared 5h rolling cap (~12M agentic tokens/mo) Floor: 12M |
RESEARCH (low) |
| Max5x | $100/mo | rollingCap: 5x Pro usage per 5-hour session weeklyAllocation: Weekly sliding window scales with 5x multiplier overage: Usage credits at API rates (Fable draws weekly limits at 50%) |
Claude Opus 5, Claude Fable 5.1, Claude Sonnet 5, Claude Haiku 4.5 | 5x Pro rolling limit (~60M agentic tokens/mo) Floor: 60M |
RESEARCH (low) |
| Max20x | $200/mo | rollingCap: 20x Pro usage per 5-hour session weeklyAllocation: Weekly sliding window scales with 20x multiplier overage: Usage credits at API rates |
Claude Opus 5, Claude Fable 5.1, Claude Sonnet 5, Claude Haiku 4.5 | 20x Pro rolling limit (~240M agentic tokens/mo) Floor: 240M |
RESEARCH (low) |
Key Gotchas & Constraints:
- Usage is quota-based per rolling 5-hour session plus weekly caps, deliberately NOT token- or message-metered
- In May 2026, Anthropic officially doubled the 5-hour rolling rate limits for Pro, Max, and Team plans and removed peak-hour throttling cliffs
- After limits, work continues only if you opt into usage credits billed at standard API rates (with optional monthly spend cap)
- Claude Code shares the same usage pool as Claude.ai chat, desktop and mobile
- On Pro, Claude Fable 5.1 runs via usage credits only; on Max it consumes 50% of weekly limits
- Cache TTL is 1 hour on subscriptions but drops to 5 minutes while drawing usage credits
- Deep-dive telemetry: see
docs/OSINT_USAGE_STATISTICS.md
Cursor (cursor)
- Primary Pricing & Docs: https://cursor.com/docs/account/pricing
- Category:
coding-ide - Data Privacy & Training: Opt-out available in Privacy Mode
- IP Indemnity: False
- Stacking Policy:
silent— "Terms of Service contain no account-sharing or multi-account clause; only resale/lease/lending of the Service is prohibited."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Hobby | Free ($0) | agentRequests: Limited Agent requests composer: Access to Composer |
Cursor Grok 4.5, Claude Haiku 4.5 | Limited Agent requests (~0.5M tokens/mo) Floor: 0.5M |
RESEARCH (low) |
| Pro | $20/mo ($192/yr) |
agentRequests: Extended Agent limits usagePools: Cursor Models pool + Other Models pool (third-party at API rate), monthly reset onDemand: Extra usage on-demand at API rates, billed in arrears |
Cursor Grok 4.6, Composer 2.5, Claude Fable 5.1 (+6 more) |
Pro Cursor-Models + Other-Models pools (~10M tokens/mo modeled) Floor: 10M |
RESEARCH (low) |
| Pro Plus | $60/mo ($576/yr) |
agentRequests: Larger usage pools usagePools: Same two pools as Pro, expanded |
Cursor Grok 4.6, Composer 2.5, Claude Fable 5.1 (+4 more) |
Official 3x Pro limits (~30M tokens/mo) Floor: 30M |
RESEARCH (low) |
| Ultra | $200/mo ($1920/yr) |
agentRequests: Largest usage pools usagePools: Same two pools, maximum size |
Cursor Grok 4.6, Composer 2.5, Claude Fable 5.1 (+3 more) |
Official 20x Pro limits (~200M tokens/mo) Floor: 200M |
RESEARCH (low) |
Key Gotchas & Constraints:
- Pricing uses two usage pools (Cursor Models vs Other Models at API rates) resetting monthly - not request counts
- On-demand usage after pool exhaustion is billed in arrears at API rates
- Max Mode and fast-request tiers exist only for grandfathered legacy subscriptions
- Teams adds a Cursor token rate ($0.25/M tokens) on third-party models
- India-only Start plan (Rs 649) limits users to the Cursor Models pool only
- Teams Standard ($40/user) and Premium ($120/user) plans are sold separately and are not modeled as tiers here
- Subscriptions are only sold directly via cursor.com; resellers are unauthorized and may be suspended
- Deep-dive telemetry: see
docs/OSINT_USAGE_STATISTICS.md
GitHub Copilot (github-copilot)
- Primary Pricing & Docs: https://github.com/features/copilot/plans
- Category:
coding-ide - Data Privacy & Training: Opt-out available (individual); excluded on Business/Enterprise
- IP Indemnity: False
- Stacking Policy:
silent— "GitHub ToS: "Your login may only be used by one person - i.e., a single login may not be shared by multiple people"; multiple paid accounts are not addressed (one free account per person)."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Free | Free ($0) | completions: 2,000 completions/mo aiCredits: Limited allowance (auto model selection only) agents: Limited |
Auto model selection only | Limited free allowance (~0.4M tokens/mo) Floor: 0.4M |
RESEARCH (low) |
| Pro | $10/mo | completions: Unlimited completions on paid tiers aiCredits: 1,000 base + 500 flex = 1,500 GitHub AI credits/mo |
Claude Sonnet 5, Claude Haiku 4.5, GPT-5.4 (+5 more) |
1,000 base AI credits/mo (~4.8M tokens; up to 1,500 with Flex) Floor: 4.8M |
Mid: 7.1M |
| Pro+ | $39/mo | completions: Unlimited completions aiCredits: 3,900 base + 3,100 flex = 7,000 GitHub AI credits/mo agents: Delegate to Claude/Codex agents |
Claude Opus 5, Claude Fable 5.1, Claude Sonnet 5 (+4 more) |
3,900 base AI credits/mo (~18.6M tokens; up to 7,000 with Flex) Floor: 18.6M |
Mid: 33M |
| Max | $100/mo | completions: Unlimited completions aiCredits: 10,000 base + 10,000 flex = 20,000 GitHub AI credits/mo |
Claude Opus 5, Claude Fable 5.1, GPT-6 Astra (+2 more) |
10,000 base AI credits/mo (~47.6M tokens; up to 20,000 with Flex) Floor: 47.6M |
Mid: 95M |
| Business | $19/mo | aiCredits: 1,900 AI credits/user/mo (org-level billing) policy: Org policy control, no model training on org content |
All premium models (per-user allowance) | 1,900 AI credits/user/mo (~9M tokens/user) Floor: 9M |
RESEARCH (low) |
| Enterprise | $39/mo | aiCredits: 3,900 AI credits/user/mo (org pool) overage: $0.01 per credit |
All premium models (org pool) | 3,900 AI credits/user/mo pooled (~19M tokens/user) Floor: 19M |
RESEARCH (low) |
Key Gotchas & Constraints:
- Premium requests were replaced by GitHub AI Credits: 1 credit = $0.01, monthly reset with a variable 'Flex' allotment on top of base credits
- Opus 5, Fable 5/5.1 and GPT-5.4 nano are Pro+/Max only; Free/Student tiers only get auto model selection
- o3-mini, o1 and GPT-4.1 were retired from Copilot in Oct 2025-2026; Opus 4.5/4.6, Sonnet 4.5 and Gemini 3.1 Pro retired Sep 1 2026
- Premium models cost multiple credits per request; task cost varies with model and complexity
- Use of interaction data to train models can be opted out on individual plans; never on Business/Enterprise
Google Antigravity (google-antigravity)
- Primary Pricing & Docs: https://antigravity.google/pricing
- Category:
coding-ide - Data Privacy & Training: Interactions used for ML improvement with in-settings opt-out; deletion on request via [email protected]
- IP Indemnity: False
- Stacking Policy:
silent— "Google consumer terms and Generative AI Additional Terms reviewed; no multi-account or account-sharing clause found. Third-party tooling against Antigravity OAuth is banned."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Individual | Free ($0) | quota: Meaningful weekly quota, refreshed weekly (no 5-hour refresh) rateLimit: Basic weekly rate limits per model tabCompletions: Unlimited commandRequests: Unlimited creditOverage: Not available (Pro/Ultra only) |
Gemini 3.8 Flash, Gemini 3.7 Flash, Gemini 3.6 Flash (+4 more) |
Weekly-refresh baseline quota (~6.3M tokens/mo) Floor: 6.25M |
Mid: 13.75M |
| Google AI Pro | $19.99/mo | quota: High quota, refreshed every 5 hours until weekly limit reached rateLimit: Higher weekly rate limit per model tabCompletions: Unlimited creditOverage: AI credits consumed at standard Gemini Enterprise Agent Platform pricing ('Never' or 'Always' toggle) |
Gemini 3.8 Flash, Gemini 3.7 Flash, Gemini 3.6 Flash (+4 more) |
5-hour refresh quota with weekly caps (~75M tokens/mo) Floor: 75.0M |
Mid: 165.0M |
| Google AI Ultra 5x | $99.99/mo | quota: Highest quota, refreshed every 5 hours until weekly limit reached rateLimit: Highest weekly rate limit per model tabCompletions: Unlimited creditOverage: AI credits consumed at standard GEAP consumption pricing |
Gemini 3.8 Flash, Gemini 3.7 Flash, Gemini 3.6 Flash (+4 more) |
Official 5x Google AI Pro limits (~375M tokens/mo) Floor: 375.0M |
Mid: 825.0M |
| Google AI Ultra 20x | $199.99/mo | quota: Highest quota, refreshed every 5 hours until weekly limit reached rateLimit: Highest weekly rate limit per model tabCompletions: Unlimited creditOverage: AI credits consumed at standard GEAP consumption pricing |
Gemini 3.8 Flash, Gemini 3.7 Flash, Gemini 3.6 Flash (+4 more) |
Official 20x Google AI Pro limits (~1.5B tokens/mo) Floor: 1500.0M |
Mid: 3300.0M |
| Organization (Google Cloud) | Enterprise | billing: Consumption-based pricing via Gemini Enterprise Agent Platform governance: Google Cloud Terms of Service; Cloud project integration exclusion: Claude and GPT-OSS reasoning models not available |
Gemini 3.8 Flash, Gemini 3.7 Flash, Gemini 3.6 Flash, Gemini 3.1 Pro | Pay-per-use at actual GEAP consumption rates - no fixed monthly token bundle Floor: 0M |
RESEARCH (low) |
Key Gotchas & Constraints:
- Quotas are NOT token- or task-metered: rate limits are 'correlated with the amount of work done by the agent', so harder tasks consume allowance faster
- No official task or token counts are published; estimatedTokenBudget figures are research estimates and low confidence
- Pro/Ultra use a 5-hour refresh until the weekly cap is hit; free tier refreshes weekly only
- AI-credit overage is billed at standard Gemini Enterprise Agent Platform consumption pricing once baseline quota is exhausted
- Using third-party software/tools against Antigravity OAuth (e.g. OpenClaw) breaches the ToS and can lead to account suspension
- No BYOK or bring-your-own-endpoint support, even on paid tiers
- Google Cloud/Enterprise usage is governed by Google Cloud terms instead of the consumer terms
- Deep-dive telemetry: see
docs/OSINT_USAGE_STATISTICS.md
Kiro (kiro)
- Primary Pricing & Docs: https://kiro.dev/pricing/
- Category:
coding-ide - Data Privacy & Training: Not published on pricing page
- IP Indemnity: False
- Stacking Policy:
silent— "FAQ: "subscriptions and usage limits are calculated per individual user"; sharing is not permitted, multiple paid accounts are not addressed."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Free | Free ($0) | credits: 50 credits/mo models: Rate-limited Sonnet-class + open-weight models |
Claude Sonnet 4.5, Qwen3 Coder Next, DeepSeek V3.2, MiniMax M2.1 | 50 credits/mo (~1M tokens) Floor: 1M |
RESEARCH (low) |
| Pro | $20/mo | credits: 1,000 credits/mo overage: Add-on credit packs available |
Claude Sonnet 5, Claude Opus 5, Auto mode (+3 more) |
1,000 credits/mo (~20M tokens on Sonnet 5) Floor: 20M |
RESEARCH (low) |
| Pro+ | $40/mo | credits: 2,000 credits/mo overage: Add-on credit packs available |
Claude Sonnet 5, Claude Opus 5, Auto mode (+3 more) |
2,000 credits/mo (~40M tokens on Sonnet 5) Floor: 40M |
RESEARCH (low) |
| Pro Max | $100/mo | credits: 5,000 credits/mo overage: Add-on credit packs available |
Claude Sonnet 5, Claude Opus 5, Auto mode, Open-weight models | 5,000 credits/mo (~100M tokens on Sonnet 5) Floor: 100M |
RESEARCH (low) |
| Power | $200/mo | credits: 10,000 credits/mo overage: Add-on credit packs available |
Claude Sonnet 5, Claude Opus 5, Auto mode, Open-weight models | 10,000 credits/mo (~200M tokens on Sonnet 5) Floor: 200M |
RESEARCH (low) |
Key Gotchas & Constraints:
- Credits reset monthly at billing-cycle start, not rolling 5-hour windows
- Credits are consumed fractionally per request; complex spec-driven tasks burn many credits
- Not all premium models are available in every country/region
- AWS owns Kiro - usage governed under AWS terms
Lovable (lovable)
- Primary Pricing & Docs: https://lovable.dev/pricing
- Category:
coding-ide - Data Privacy & Training: Excluded by default on Business+; opt-out available on Free/Pro
- IP Indemnity: False
- Stacking Policy:
silent— "Terms ban sharing credentials; workspaces support unlimited members and are priced by credits, not seats - multiple paid subscriptions on one account are not addressed."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Free | Free ($0) | buildCredits: 5 daily build credits (cap 30/mo) grants: 20 credits/mo Cloud grant + 4 credits/mo AI grant |
Lovable Agent | 30 build credits + monthly grants (~3M tokens/mo) Floor: 3M |
RESEARCH (low) |
| Pro 100 | $25/mo | buildCredits: 100 credits/mo (annual $21/mo) dailyFloor: 5 daily build credits + 20 Cloud + 4 AI grants stack on top rollover: Monthly credits roll over (expire 2 months later on monthly billing) |
Lovable Agent | 100+ credits/mo (~15M tokens) Floor: 15M |
RESEARCH (low) |
| Business | $50/mo | buildCredits: 100 credits tier at $50/mo (annual $42) - scales with tier privacy: Workspace data excluded from AI model training by default |
Lovable Agent | Business minimum tier (~40M tokens/mo at 200+ cr tiers typical) Floor: 40M |
RESEARCH (low) |
Key Gotchas & Constraints:
- Credits are a unified balance covering app builds, hosting (Cloud) and AI gateway usage of deployed apps
- Free-tier credits do NOT roll over; Pro credits expire 2 months after issue on monthly plans
- Daily 5 build credits are a floor, not a bonus: monthly credits also include daily research
- Higher credit tiers scale steeply: 200 cr $50, 400 cr $100, 800 cr $200 ... 10,000 cr $2,250
Meta Muse Code (meta-muse-code)
- Primary Pricing & Docs: https://developer.meta.com/ai/products/muse-code
- Category:
coding-ide - Data Privacy & Training: Depends on the selected models and API tier; Muse Code keys are limited to Muse Code and training terms are not uniform across models
- IP Indemnity: False
- Stacking Policy:
unknown
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Everyday | $5/mo | prompts: 10-50 prompts / 5h (official) | Muse Spark 1.3, Llama 4 Scout | Light usage tier (~5M tokens/mo) Floor: 5M |
RESEARCH (low) |
| High | $15/mo | prompts: 5x Everyday usage (official) | Muse Spark 1.3, Llama 4 Scout, Llama 4 Maverick | 5x Everyday usage (~25M tokens/mo, official multiplier) Floor: 25M |
RESEARCH (low) |
| Power | $50/mo | prompts: 20x Everyday usage (official) | Muse Spark 1.3, Llama 4 Maverick | 20x Everyday usage (~100M tokens/mo, official multiplier) Floor: 100M |
RESEARCH (low) |
Key Gotchas & Constraints:
- Muse Code subscription keys work only with Muse Code, not the general Model API
- Training terms depend on which models you select; the Contributor/Standard choice changes them
- Regional availability restrictions apply
OpenAI Codex (ChatGPT) (openai-codex)
- Primary Pricing & Docs: https://openai.com/chatgpt/pricing/
- Category:
coding-ide - Data Privacy & Training: Used for training; opt-out available
- IP Indemnity: False
- Stacking Policy:
silent— "Terms ban sharing account credentials; multiple paid accounts are not addressed."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Free | Free ($0) | codex: Limited Codex access | GPT-5.6 Luna | Limited Codex (~0.75M tokens/mo) Floor: 0.75M |
Mid: 1.65M |
| Go | $8/mo | codex: Limited Codex access; limited GPT-5.6 Terra bonus: More uploads/voice/deep research, longer memory; may include ads |
GPT-5.6 Luna, GPT-5.6 Terra | Light coding assistance (~4.3M tokens/mo) Floor: 4.25M |
Mid: 9.35M |
| Plus | $20/mo | codex: Expanded Codex usage context: GPT Reasoning 256K |
GPT-6 Astra, GPT-5.6 Sol, GPT-5.6 Terra | Expanded Codex (~20M tokens/mo) Floor: 20.0M |
Mid: 44.0M |
| Pro (5x) | $100/mo | codex: 5x Plus usage; high-volume Codex | GPT-6 Astra, GPT-5.6 Sol | 5x Plus (~100M tokens/mo) Floor: 100.0M |
Mid: 220.0M |
| Pro (maximum) | $200/mo | codex: Maximum Codex tasks context: GPT Reasoning 400K context |
GPT-6 Astra, GPT-5.6 Sol Pro | Maximum Codex (~400M tokens/mo) Floor: 400.0M |
Mid: 880.0M |
Key Gotchas & Constraints:
- Codex ships inside ChatGPT plans, not a separate product
- Codex access is Limited on Free/Go, Expanded on Plus, Maximum on Pro
- GPT-5.3-Codex appears in other registries (GitHub Copilot) but is not marketed on ChatGPT pricing
- Content is used to train OpenAI models; opt-out available on all tiers
- Deep-dive telemetry: see
docs/OSINT_USAGE_STATISTICS.md
Replit (replit)
- Primary Pricing & Docs: https://replit.com/pricing
- Category:
coding-ide - Data Privacy & Training: Not published on pricing page
- IP Indemnity: False
- Stacking Policy:
prohibited— "Terms: "Creating accounts with automation or registering multiple accounts" is prohibited."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Core | $20/mo ($216/yr) |
agentHours: Up to 30 hours of chat in Free Mode projects: Up to 60 projects on Free Mode modelCredits: $20 toward most powerful models |
Claude Opus 5, GPT-5.6 Sol, DeepSeek V4-Pro, Gemini 3.8 Flash, Replit Agent | $20 credit pool + 30h (~24M baseline; 2.09M Opus 5 to 287.18M Gemini Flash) Floor: 24M |
RESEARCH (low) |
| Pro | $100/mo ($1080/yr) |
parallelAgents: 10 parallel agents modelCredits: $100 toward most powerful models collaboration: 15 collaborators, 50 viewers |
Claude Opus 5, GPT-5.6 Sol, DeepSeek V4-Pro, Gemini 3.8 Flash, Replit Agent | $100 credit pool (~120M baseline; 10.47M Opus 5 to 1435.9M Gemini Flash) Floor: 120M |
RESEARCH (low) |
| Enterprise | Enterprise | billing: Custom seats; SSO/SAML, single-tenant, static IPs | Custom fleet | Custom contract Floor: 0M |
RESEARCH (low) |
Key Gotchas & Constraints:
- $20/$100 'toward most powerful models' is a dollar credit pool spent at list prices across Claude Opus 5 (2.09M/10.47M), GPT-5.6 Sol (7.85M/39.25M), DeepSeek V4-Pro (60.54M/302.7M), Gemini 3.8 Flash (287.18M/1435.9M), and Replit Agent
- Beyond included Free-Mode hours, effort-based pricing applies (agent-hours billed as usage)
- Optional Prepacks: $90/$215/$425/$825/$2,000 per month
- Exact checkpoint consumption per tier not officially published
- Stacking prohibited: multi-account creation banned in terms; clamped to 1 copy in Dangerous Dave mode
Tabnine (tabnine)
- Primary Pricing & Docs: https://www.tabnine.com/pricing/
- Category:
coding-ide - Data Privacy & Training: Training never occurs on customer code
- IP Indemnity: True
- Stacking Policy:
silent— "Terms ban sharing an Account/Login and sell seats per registered user; multiple paid accounts are not addressed."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Code Assistant | $39/mo | billing: $39/user/mo (annual subscription) usage: Unlimited when using your own LLM on-prem/VPC; Tabnine-provided LLM access billed at provider list price + 5% handling fee deployment: SaaS, VPC, on-prem, or air-gapped; zero code retention |
Tabnine Protected | Platform fee; ~19.5M fee-equivalent tokens Floor: 19.5M |
Mid: 39M |
| Agentic Platform | $59/mo | billing: $59/user/mo (annual subscription) usage: Same LLM economics as Code Assistant included: Context Engine, Tabnine CLI agents, MCP tool use, headless agents (optional add-on) |
Tabnine Protected | Platform fee; ~29.5M fee-equivalent tokens Floor: 29.5M |
Mid: 59M |
| Enterprise (Custom) | Enterprise | volume: Quote-based; IP indemnification subject to terms compliance: GDPR, SOC 2, ISO 27001 |
Tabnine Protected | Custom volume contract Floor: 0M |
RESEARCH (low) |
Key Gotchas & Constraints:
- Tabnine was acquired by Tricentis (agentic quality engineering) and consumer self-serve tiers (Starter/Pro $15) are discontinued
- Pricing is enterprise-first: annual per-user subscription with quote-based procurement
- Using your own LLM endpoint = unlimited usage; Tabnine-provided LLM access adds a 5% handling fee over provider list prices
- Reserve-token model means monthly token totals are not fixed
Windsurf (Cognition) (windsurf)
- Primary Pricing & Docs: https://windsurf.com/pricing
- Category:
coding-ide - Data Privacy & Training: Not published on pricing page
- IP Indemnity: False
- Stacking Policy:
silent— "Acceptable Use Policy bans "sharing credentials"; multiple paid accounts are not addressed."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Free | Free ($0) | quota: Light auto-refreshing usage allowance (daily + weekly) models: Limited model availability concurrency: Up to 10 concurrent sessions |
SWE-2, Kimi K2.5, GPT 5.2 Mini, Claude Haiku 4.5 | Light agent quota (~3.3M tokens/mo) Floor: 3.25M |
Mid: 7.15M |
| Pro | $20/mo | quota: Increased auto-refreshing daily + weekly allowance (size not published) overage: Extra usage purchased at API pricing concurrency: Up to 10 concurrent sessions |
SWE-2, Claude Sonnet 5, Claude Haiku 4.5 (+2 more) |
Pro daily+weekly quota (~75M tokens/mo) Floor: 75.0M |
Mid: 165.0M |
| Max | $200/mo | quota: Significantly higher daily + weekly allowance (size not published) overage: Extra usage at API pricing concurrency: Unlimited concurrent sessions |
SWE-2, Claude Sonnet 5, Claude Haiku 4.5 (+2 more) |
Max daily+weekly quota (~375M tokens/mo) Floor: 375.0M |
Mid: 825.0M |
| Teams | $80/mo | quota: Pro-level quota per full dev seat billing: $80/mo team base + $40/mo per full user seat concurrency: Unlimited concurrent sessions |
SWE-2, Claude Sonnet 5, Claude Haiku 4.5 (+2 more) |
Pro-level quota per full user seat (~75M tokens/mo/seat) Floor: 75.0M |
Mid: 165.0M |
Key Gotchas & Constraints:
- After Cognition's acquisition, windsurf.com/pricing serves the Devin-branded plan sheet; there is no separate Windsurf editor pricing
- Quotas are undisclosed 'usage allowances' that refresh daily and weekly; overage is consumed at API pricing
- SWE-2 is included free in Devin Desktop and CLI only through Oct 10, 2026
- Message cost varies by model, task size and complexity - not a fixed token count
- Old 'Cascade prompts' / 'Fast prompts' terminology is retired from pricing
- On March 19, 2026, Windsurf replaced credit blocks with auto-refreshing daily and weekly usage quotas deeply integrated with Devin
- Deep-dive telemetry: see
docs/OSINT_USAGE_STATISTICS.md
2. Coding Routers & Aggregators (5 Services)
CommandCode (commandcode)
- Primary Pricing & Docs: https://commandcode.ai/pricing
- Official Usage Limits & Rolling Caps: https://commandcode.ai/docs/resources/usage-limits
- Usage Estimates & Request Sizing: https://commandcode.ai/docs/resources/pricing-limits#usage-estimates
- Go Plan Specification: https://commandcode.ai/docs/plans/go
- GOAT Plan Specification & Model Allowances: https://commandcode.ai/docs/plans/goat
- Pro Plan Specification & Premium Allowances: https://commandcode.ai/docs/plans/pro
- Max Plans & Dual-Pool Credit Rules: https://commandcode.ai/docs/plans/max#usage-limits · https://commandcode.ai/docs/plans/max#how-credits-work
- Category:
coding-router - Data Privacy & Training: No training on your code (Zero Data Retention available via
CMD_ZDR=1) - IP Indemnity: False
- Stacking Policy:
prohibited— "Terms: "One account per person. You may not register, operate, or control more than one account... Creating, operating, or controlling multiple accounts is a material breach.""
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Go | $1/mo | monthlyCredits: $10/mo compute credits 5h cap: $3 rolling limit weekly cap: $6 rolling limit requests: ~15K mix (~26K DeepSeek V4 Flash at typical cache) |
DeepSeek V4 Flash, GPT-5.6 Luna, GLM-5.3 Flash, MiniMax M3, Qwen 3.8 Max, Grok 4.5 | $10 compute credits (~15M tokens on open models) Floor: 15M · Mid: 26M · Opt: 84M |
OFFICIAL (high) "On Go, DeepSeek V4 Flash runs ~ 42K requests with no cache - ~ 26K once the typical 50K cache reads are included." |
| GOAT | $10/mo | monthlyCredits: $70/mo usage value credits 5h cap: $14 rolling limit weekly cap: $35 rolling limit requests: ~75K mix (~154K with DeepSeek V4 Flash) allowances: $70 Sol/GLM-5.2/Hy3 · $60 DeepSeek Flash · $47 MiniMax M3 · $40 GLM Flash · $20 Luna/Grok |
DeepSeek V4 Flash, GPT-5.6 Sol, GLM-5.2, GLM-5.3 Flash, Tencent Hy3, Qwen 3.8 27B, MiniMax M3, Gemini 3.8 Flash, GPT-5.6 Luna, Grok 4.5 | $70 compute credits (~70M tokens) Floor: 70M · Mid: 120M · Opt: 751M |
OFFICIAL (high) "You pay $10 and code with up to $70 of per-model credit allowances... $14 of usage in any 5 hours, $35 in any 7 days, and $70 per month." |
| Pro | $20/mo | monthlyCredits: $80/mo usage value credits 5h cap: $16 rolling limit weekly cap: $40 rolling limit requests: ~100K mix (~47% standard / ~53% premium) allowances: $80 Sol/GLM-5.2/Hy3 · $70 DeepSeek Flash · $57 MiniMax M3 · $50 GLM Flash · $20 Claude Sonnet/Opus |
Claude Sonnet 5, Claude Opus 4.8, GPT-5.6 Sol, DeepSeek V4 Flash, GLM-5.2, GLM-5.3 Flash, Gemini 3.8 Flash, MiniMax M3, GPT-5.6 Terra | $80 compute credits (~80M tokens) Floor: 80M · Mid: 120M · Opt: 877M |
OFFICIAL (high) "Pro works the same way: 16 credits in any 5 hours and 40 in any 7 days across its per-model credits - premium models included - worth $16 / $40 of usage on a full-allowance model." |
| Max 10x | $100/mo | standardPool: $150/mo standard model usage limit premiumPool: $100/mo premium model usage limit 5h cap: $45 rolling limit weekly cap: $90 rolling limit requests: ~230K mix (~43% standard / ~57% premium) |
Claude Sonnet 5, Claude Opus 4.8, GPT-5.6 Sol, DeepSeek V4 Flash, GLM-5.3 Flash, GPT-5.6 Terra, MiniMax M3, Grok 4.5 | $150 standard + $100 premium credits Floor: 150M · Mid: 230M · Opt: 500M |
OFFICIAL (high) "On Max 10×, you have a $150 standard model usage limit and a $100 premium model usage limit... $45 of usage in any 5 hours and $90 in any 7 days." |
| Max 20x | $200/mo | standardPool: $300/mo standard model usage limit premiumPool: $200/mo premium model usage limit 5h cap: $90 rolling limit weekly cap: $180 rolling limit requests: ~370K mix (~43% standard / ~57% premium) |
Claude Sonnet 5, Claude Opus 4.8, GPT-5.6 Sol, DeepSeek V4 Flash, GLM-5.3 Flash, GPT-5.6 Terra, MiniMax M3, Grok 4.5 | $300 standard + $200 premium credits Floor: 300M · Mid: 370M · Opt: 1000M |
OFFICIAL (high) "On Max 20×, that's $300 standard and $200 premium... $90 of usage in any 5 hours and $180 in any 7 days." |
Key Gotchas & Constraints:
- Vendor request definition: ~700–1K input tokens, ~125–200 output tokens, ~50K average prompt cache reads
- Paced by two rolling windows (5-hour and weekly) that start on first use; extra on-demand top-up credits bypass rolling caps
- Max 10x and Max 20x enforce distinct standard vs. premium model usage limits ($150/$100 on Max 10x, $300/$200 on Max 20x). Sol, Luna, Grok, and Gemini bill as standard; Claude Opus, Sonnet, and Fable bill as premium
- Domain moved from commandcode.dev to commandcode.ai
- Every plan adds a small processing fee on top of the sticker price
- Credits roll over forever; auto top-up billed at API cost
- Free tier retired - now paid ($1 Go tier); taste-1 learning data stored locally only
Kilo Code (kilo-code)
- Primary Pricing & Docs: https://kilo.ai/pricing
- Category:
coding-router - Data Privacy & Training: No training on your code (official)
- IP Indemnity: False
- Stacking Policy:
silent— "Per-seat Teams licensing; terms ban credential sharing but no multiple-account clause was found."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Individual | Free ($0) | models: Access to 500+ models via Kilo Gateway or BYOK gatewayFee: 5% on all plans (no markup vs list price) |
Kilo Gateway (500+ models) | PAYG at provider list rates + 5% gateway fee Floor: 0M |
RESEARCH (low) |
| Teams | $15/mo | seats: $15/user/mo collaboration platform inference: Billed at provider rates; EU inference available markup: Inference passed through at provider rates with no markup; credit top-ups carry a 5% fee kiloPass: Kilo Pass Starter $19 / Pro $49 / Expert $199; up to 50% bonus credits (annual: 50% every month) |
Kilo Gateway (500+ models) | Platform fee; ~15M fee-equivalent tokens at open-model rates Floor: 15M |
Mid: 30M |
| Enterprise | Enterprise | volume: Quote-based support: Dedicated |
Kilo Gateway (500+ models) | Custom agree - quote-based Floor: 0M |
RESEARCH (low) |
Key Gotchas & Constraints:
- Rebranded kilocode.ai -> kilo.ai; pricing restructured to platform fee + inference + cloud compute
- Kilo Pass converts dollars 1:1 into paid credits, then adds bonus credits (50% month one; streak bonuses on monthly, 50% every month on annual)
- Bonus credits expire monthly; paid credits apply across IDE/CLI/Cloud Agents/Gateway
- Cloud agents billed per second (Gas Town $1.20/hr, Code Review $0.33/hr, Cloud Agent $0.60-1.20/hr)
Kimi Code (kimi-code)
- Primary Pricing & Docs: https://www.kimi.com/coding-plan/
- Category:
coding-router - Data Privacy & Training: Not published on pricing page
- IP Indemnity: False
- Stacking Policy:
prohibited— "User agreement: "not register or operate multiple accounts for abusive purposes" and "You may not share your account credentials or make your account available to anyone else.""
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Moderato | $19/mo ($180/yr) |
quota: Weekly-refreshed usage quota (~300 requests / 5h windows) models: Kimi K3 incl. K3-256k |
Kimi K3, K2.7 Code | Weekly-refreshed quota (~60M tokens/mo) Floor: 60M |
RESEARCH (low) |
| Allegretto | $39/mo ($372/yr) |
quota: Higher weekly limits + higher concurrency speed: K3 HighSpeed (5-6x speed, ~3x credit burn); 1M-context K3 |
Kimi K3, K2.7 Code | Higher weekly quota (~150M tokens/mo) Floor: 150M |
RESEARCH (low) |
| Allegro | $99/mo ($948/yr) |
quota: Expansive quota (~600-800 req/5h) concurrency: Higher caps |
Kimi K3, K2.7 Code | Expansive quota (~400M tokens/mo) Floor: 400M |
RESEARCH (low) |
| Vivace | $199/mo ($1908/yr) |
quota: Highest weekly quotas (~1,200 req/5h) concurrency: Max caps |
Kimi K3, K2.7 Code | Highest quota (~800M tokens/mo) Floor: 800M |
RESEARCH (low) |
Key Gotchas & Constraints:
- Quotas refresh weekly with ~5-hour rolling request windows (roughly 300-1,200 req/window by tier)
- K3 HighSpeed runs 5-6x faster but burns ~3x credits
- Exact credit counts per window are not officially published - estimates are research-derived
- K2.7 Code is the coding-optimized workhorse; K3 is the general flagship
OpenCode (opencode)
- Primary Pricing & Docs: https://opencode.ai/go · Usage Limits: https://opencode.ai/docs/go/#usage-limits · Zen PAYG: https://opencode.ai/zen
- Category:
coding-router - Data Privacy & Training: Zero data retention (0 days) on most Go models; 30 days on Grok 4.6 & GPT-5.6 Luna; Muse Spark Contributor trains on code. Zen models US-hosted with zero retention
- IP Indemnity: False
- Stacking Policy:
prohibited— "Terms of Service: prohibits users who "create, maintain, or use multiple accounts to circumvent usage limits, access restrictions, billing obligations... or any other restriction or policy.""
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| OpenSource CLI | Free ($0) | usage: Free open-source agent; BYOK to any provider | BYOK any provider | Token cost = your provider's rates Floor: 0M |
RESEARCH (low) |
| Zen (PAYG) | $20/mo | billing: Pay-as-you-go dollar balance, min $20 top-up (+$1.23 card fee) markup: Zero markup per request autoTopUp: $20 when balance < $5 |
GPT-5.6 Luna, GPT-5.6 Sol, Claude Fable 5.1 (+5 more) |
Prepaid balance (~10M fee-equivalent tokens) Floor: 10M |
Mid: 20M |
| Go | $10/mo | monthlyAllowance: Per-model monthly usage: $60 (tier 1 models), $30 (tier 2), $15 (tier 3) fiveHourCap: 20% of monthly allowance in any 5h ($12 / $6 / $3) weeklyCap: 50% of monthly allowance in any 7d ($30 / $15 / $7.50) monthlyCap: 100% of monthly allowance ($60 / $30 / $15) requests: Up to 31,580/mo on GLM-5.3-Flash, 65,000/mo on DeepSeek Flash, 150,400/mo on MiMo-V2.5, 226,600/mo on Muse Spark topUp: Optional 'Use balance' falls back to Zen PAYG balance when limits reached |
Kimi K3, Qwen 3.8 Max, Grok 4.6, GLM-5.3, GLM-5.2, DeepSeek V4.1 Flash, MiniMax M3, MiMo-V2.5, Muse Spark (28 models) |
$10/mo open-model subscription ($15–$60 per-model allowances; ~30M–752M tokens) Floor: 30M |
Mid: 95M |
Key Gotchas & Constraints:
- OpenCode Go ($10/mo) provides $15, $30, or $60 monthly usage allowance per model with 20% 5h and 50% weekly caps
- Optional 'Use balance' falls back to Zen PAYG balance when Go model limits are reached instead of blocking requests
- DeepSeek V4.1 Flash has 4× promotional limits ($60/mo limit; 26K req / 5h) through Sep 20, 2026 (normally $15)
- Muse Spark Contributor models train on prompt/output data and are not zero data retention (ZDR); all other Go models provide zero training (0 days retention, except Grok 4.6 and GPT-5.6 Luna with 30 days retention)
- Zen PAYG charges $1.23 card processing fee on top-ups; auto top-up at $20 below $5 balance
- The CLI itself is free/open-source; opencode.ai/pricing is retired (404) - offerings are Go ($10/mo) and Zen (PAYG)
OpenRouter (openrouter)
- Primary Pricing & Docs: https://openrouter.ai
- Category:
coding-router - Data Privacy & Training: No (OpenRouter does not train; providers vary)
- IP Indemnity: False
- Stacking Policy:
prohibited— "Terms: "create multiple accounts as a single user, for purposes of bypassing or circumventing use limits... or for any other reason" is prohibited."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Free | Free ($0) | rateLimit: 20 RPM dailyCap: 50-1,000 RPD (50 RPD without payment history) |
Free tier models (:free tag) | 50-1,000 requests/day on free models (~2M tokens/mo) Floor: 2M |
RESEARCH (low) |
| PAYG | Enterprise | platformFee: 5.5% on top-ups rateLimit: Provider-native limits concurrency: Provider-dependent |
Over 400+ models from 30+ providers | Wholesale PAYG ($10 buys ~5M–66M tokens depending on model) Floor: 20M |
RESEARCH (low) |
Key Gotchas & Constraints:
- Free tier restricted to 50 requests/day unless account has payment history
- 5.5% platform fee charged on crypto and card balance reloads
- Provider fallback order may route to higher-cost endpoints if not pinned
3. Direct APIs & Cloud Token Plans (15 Services)
Alibaba Cloud (alibaba-cloud)
- Primary Pricing & Docs: https://www.alibabacloud.com/help/en/model-studio/token-plan-overview
- Category:
api-provider - Data Privacy & Training: Personal: governed by the service agreement with cross-border transfer, no no-training commitment; Team: conversation data not used for model training
- IP Indemnity: False
- Stacking Policy:
prohibited— "Token Plan FAQ: "Can multiple people share one account? No... sharing the same account or API Key among multiple people is not allowed"; Team seats are bound to one member and one API key."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Personal Lite | $6/mo | quota: 2,500 credits per 7-day window (official; original price $8/mo) window: 7-day timer starts at first call; unused quota expires, service pauses when exhausted concurrency: 1-2 concurrent agents extraBundle: $15 = 20,000 credits, requires active subscription (max 5), not subject to the 7-day window region: Singapore region only |
Qwen 3.7 Plus, Qwen 3.6 Plus, Kimi K2.5 (+2 more) |
2,500 credits/7d; ~45M conservative (qwen3.6-plus basis, derived) Floor: 45M |
Mid: 90M |
| Essential | $10/mo | quota: 5,625 credits per 7-day window (official; original price $16/mo) window: 7-day timer starts at first call; unused quota expires concurrency: 2-3 concurrent agents extraBundle: $15 = 20,000 credits, requires active subscription (max 5) region: Singapore region only |
Qwen 3.7 Plus, Qwen 3.6 Plus, Kimi K2.5 (+2 more) |
5,625 credits/7d; ~101M conservative (qwen3.6-plus basis, derived) Floor: 101M |
Mid: 202M |
| Standard | $18/mo | quota: 10,000 credits per 7-day window (official; original price $25/mo) window: 7-day timer starts at first call; unused quota expires concurrency: 3-4 concurrent agents extraBundle: $15 = 20,000 credits, requires active subscription (max 5) region: Singapore region only |
Qwen 3.7 Plus, Qwen 3.6 Plus, Kimi K2.5 (+2 more) |
10,000 credits/7d; ~180M conservative (qwen3.6-plus basis, derived) Floor: 180M |
Mid: 359M |
| Pro | $68/mo | quota: 40,000 credits per 7-day window (official; original price $80/mo) window: 7-day timer starts at first call; unused quota expires concurrency: 6-8 concurrent agents extraBundle: $15 = 20,000 credits, requires active subscription (max 5) region: Singapore region only |
Qwen 3.7 Plus, Qwen 3.6 Plus, Kimi K2.5 (+2 more) |
40,000 credits/7d; ~719M conservative (qwen3.6-plus basis, derived) Floor: 719M |
Mid: 1438M |
| Team Standard | $20/mo | quota: 25,000 credits/seat/month (official; original price $30/seat) window: Monthly batch granted at subscription start; no 7-day window; unused expires concurrency: Shared workspace; each seat bound to one member and one API key extraPack: $700 = 625,000 shared credits (soonest-expiring first) |
Qwen 3.7 Plus, Qwen 3.6 Plus, Kimi K2.5, GLM-5 | 25,000 credits/seat/mo; ~105M conservative/seat (derived) Floor: 105M |
Mid: 210M |
| Team Pro | $75/mo | quota: 100,000 credits/seat/month (official; original price $100/seat) window: Monthly batch; unused expires concurrency: Shared workspace; one API key per seat extraPack: $700 = 625,000 shared credits |
Qwen 3.7 Plus, Qwen 3.6 Plus, Kimi K2.5, GLM-5 | 100,000 credits/seat/mo; ~419M conservative/seat (derived) Floor: 419M |
Mid: 839M |
| Team Max | $200/mo | quota: 250,000 credits/seat/month (official) window: Monthly batch; unused expires concurrency: Shared workspace; one API key per seat extraPack: $700 = 625,000 shared credits |
Qwen 3.7 Plus, Qwen 3.6 Plus, Kimi K2.5, GLM-5 | 250,000 credits/seat/mo; ~1.05B conservative/seat (derived) Floor: 1048M |
Mid: 2097M |
Key Gotchas & Constraints:
- Token Plan and Coding Plan are separate products: Coding Plan (request-metered, Pro $50, 6,000 req/5h + 45,000/week + 90,000/month) is being phased out and is no longer available once sold out
- Credits are deducted with per-model coefficients that Alibaba does not publish; estimates are derived from the official qwen3.6-plus worked example
- Personal tier quotas run on a fixed 7-day window from first use and unused quota expires; calls block when exhausted (no pay-as-you-go)
- Personal plans offer no guarantee against data training and use cross-border (Singapore) processing; Team plans commit to no training on conversation data
- Team seats are bound to one member and one API key and cannot be shared
- Personal Edition is only available in the Singapore region
Anthropic API (anthropic-api)
- Primary Pricing & Docs: https://console.anthropic.com
- Category:
api-provider - Data Privacy & Training: No (Commercial API zero retention)
- IP Indemnity: True
- Stacking Policy:
silent— "Consumer/commercial terms ban sharing account credentials and API keys but do not address one person holding multiple paid accounts."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| PAYG | Enterprise | spendTiers: Start $500/mo, Build $1,000/mo, Scale $200,000/mo (official 2026 tiers) cacheReadDiscount: 90% off cached input cacheWritePremium: 25% premium on cache writes batchDiscount: 50% off |
Claude Fable 5.1, Claude Opus 5, Claude Sonnet 5, Claude Haiku 4.5 | Direct API PAYG ($20 buys ~10M conservative / ~18M cached-agent Sonnet 5 tokens) Floor: 10M |
Mid: 18M |
Key Gotchas & Constraints:
- Rate limits and spend caps are tiered by cumulative spend (Start $500, Build $1,000, Scale $200,000+/mo); new accounts hit 429s on large codebases
- Thinking/reasoning tokens are billed at standard output token rates
- Cache write operations cost 25% more than base input tokens
BytePlus ModelArk Coding Plan (byteplus)
- Primary Pricing & Docs: https://www.byteplus.com/en/activity/arkcodingplan
- Category:
api-provider - Data Privacy & Training: No training on Customer Data under BytePlus AI Services terms (except volunteered feedback)
- IP Indemnity: False
- Stacking Policy:
silent— "BytePlus AI terms cover data training but contain no multi-account clause; plans are "primarily intended for individual developers" and non-coding use may be treated as abuse."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Lite | $10/mo | usage: 3x the usage of the Claude Pro plan requests: ≈1,900 requests/5h, 12,000/week, 24,000/month (official docs; older FAQ quoted 1,200/9,000/18,000) tools: Claude Code, Cursor, Cline, Kilo Code, Roo Code, OpenCode |
Dola-Seed-2.0-Pro, Dola-Seed-2.0-Lite, GLM-5.3-Flash (+4 more) |
3x Claude Pro usage (~36M tokens/mo, anchored to our Claude Pro estimate) Floor: 36M |
DERIVED (medium) Basis: Claude Pro conservative estimate (12M) "The Lite plan provides a usable total quota equivalent to 3x that of the Claude Pro plan, while the Pro plan offers 5..." |
| Pro | $50/mo | usage: 5x Lite usage (also marketed as 3x Claude Max usage) requests: 5x Lite quotas (≈9,500/5h, 60,000/week, 120,000/month) bonus: ArkClaw included during subscription |
Dola-Seed-2.0-Pro, Dola-Seed-2.0-Lite, GLM-5.3-Flash (+3 more) |
5x Lite (~180M tokens/mo) Floor: 180M |
RESEARCH (low) |
Key Gotchas & Constraints:
- 2026 refresh renamed models from Bytedance-Seed naming to Dola-Seed-2.0 pro/lite
- Quota is marketed as a multiplier of Claude Pro/Max plans, not absolute credits
- Non-coding use of the plan may be treated as abuse and can lead to subscription deactivation or account suspension
- Model availability varies by region
DeepSeek API (deepseek-api)
- Primary Pricing & Docs: https://api-docs.deepseek.com/quick_start/pricing
- Category:
api-provider - Data Privacy & Training: No
- IP Indemnity: False
- Stacking Policy:
unknown
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| PAYG (Flash) | Enterprise | concurrency: 2,500 connections (official limit) contextWindow: 1M tokens context window, 384K maximum output offPeakDiscount: 50% off (off-peak input $0.15/M, output $0.60/M) peakHours: Mon-Fri 01:00-04:00 and 06:00-10:00 UTC |
DeepSeek V4.1 Flash | PAYG ($20 buys ~75M Flash tokens off-peak) Floor: 75M |
RESEARCH (low) |
| PAYG (V4-Pro) | Enterprise | concurrency: 500 connections (official limit) contextWindow: 1M tokens context window, 384K maximum output offPeakDiscount: 50% off off-peak vs peak |
DeepSeek V4-Pro | PAYG ($20 buys ~8M Pro tokens off-peak) Floor: 8M |
Mid: 20M |
Key Gotchas & Constraints:
- Peak hours are Mon-Fri 01:00-04:00 and 06:00-10:00 UTC (Asiapost-hours); other times 50% off
- Cache-hit tokens are nearly free but require identical prompt prefixes and are ephemeral
- $0.27-1.68/M price brackets quoted in older docs are stale; rely on current api-docs pricing
Fireworks.ai (fireworks-ai)
- Primary Pricing & Docs: https://fireworks.ai/pricing
- Category:
api-provider - Data Privacy & Training: No
- IP Indemnity: False
- Stacking Policy:
unknown
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Free | Free ($0) | trial: $1 free credits (official) freeRpm: Exact free-tier RPM unpublished; account-wide 6,000 RPM ceiling applies once a card is on file |
Llama 3.3 70B, DeepSeek V3 | $1 free trial (~2M tokens) Floor: 2M |
RESEARCH (low) |
| PAYG | Enterprise | RPM: 6,000 requests/min with valid card batchDiscount: 50% off Batch API latency: Sub-100ms TTFT |
Llama 4 Maverick, DeepSeek V4, Kimi K3 | Serverless PAYG ($20 buys ~35M tokens) Floor: 35M |
RESEARCH (low) |
Key Gotchas & Constraints:
- 503 load shedding errors occur during saturated cluster load events
- Card required on file to unlock 6,000 RPM high throughput
- Fine-tuning deployment incurs hourly standby charges
Google AI Studio (google-ai-studio)
- Primary Pricing & Docs: https://aistudio.google.com
- Category:
api-provider - Data Privacy & Training: Yes on Free tier; No on Paid tiers
- IP Indemnity: False
- Stacking Policy:
unknown
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Free | Free ($0) | RPM: 15 requests/min TPM: 1,000,000 tokens/min RPD: 1,500 requests/day |
Gemini 3.8 Flash, Gemini 3.1 Pro | 1,500 requests/day (~15M tokens/mo) Floor: 15M |
RESEARCH (low) |
| Tier1 | Enterprise | spendCap: $250/mo spend limit RPM: 1,000 requests/min TPM: 4,000,000 tokens/min |
Gemini 3.8 Flash, Gemini 3.1 Pro | PAYG up to $250/mo spend cap (~67M conservative / ~130M cached-agent tokens @ $100) Floor: 67M |
Mid: 100M |
| Tier2 | Enterprise | spendCap: $2,000/mo spend limit RPM: 2,000 requests/min TPM: 8,000,000 tokens/min |
Gemini 3.8 Flash, Gemini 3.1 Pro | PAYG up to $2,000/mo limit (~500M tokens) Floor: 500M |
RESEARCH (low) |
| Tier3 | Enterprise | spendCap: $20,000-$100,000/mo spend limit RPM: 5,000+ requests/min TPM: 20,000,000+ tokens/min |
Gemini 3.8 Flash, Gemini 3.1 Pro | Enterprise scale PAYG (~2B tokens) Floor: 2000M |
RESEARCH (low) |
Key Gotchas & Constraints:
- Free tier data IS logged and used for model training and human review
- Promotional credits expire after 1 year and are non-refundable
- Context caching incurs continuous storage fees per hour
Groq API (groq-api)
- Primary Pricing & Docs: https://console.groq.com
- Category:
api-provider - Data Privacy & Training: No
- IP Indemnity: False
- Stacking Policy:
unknown
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Free | Free ($0) | limits: Developer-plan base limits (Free is lower): per-model RPM 10-30, RPD 100-1,000, TPM 1.2K-70K, TPD up to ~500K | GPT-OSS-120B, Qwen3.8, Kimi K3 | Per-model rate caps (~2-10M tokens/mo depending on model) Floor: 5M |
RESEARCH (low) |
| PAYG | Enterprise | batchDiscount: 50% batch/caching discount throughput: 400-800 tokens/sec on LPU hardware |
GPT-OSS-120B, Qwen3.8, Kimi K3, Compound models | Direct PAYG ($20 buys ~40M tokens) Floor: 40M |
RESEARCH (low) |
Key Gotchas & Constraints:
- Llama chat/vision models are GONE from Groq's catalog (only Llama Prompt-Guard safety models remain); lineup is now GPT-OSS/Qwen/Whisper/Compound
- Free-plan limits are per-model, not a single global quota; console tables show Developer-plan base limits unless stated otherwise
- Set manual spend caps to prevent runaway loops on ultra-fast inference
Meta Llama API (meta-model-api)
- Primary Pricing & Docs: https://dev.meta.ai/docs/pricing-rate-limits
- Category:
api-provider - Data Privacy & Training: Yes for Contributor tier; No for Standard tier
- IP Indemnity: False
- Stacking Policy:
unknown
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Standard | Enterprise | RPM: 3,000 requests/min TPM: 500,000 tokens/min price: $1.25/M input, $4.25/M output (official) |
Muse Spark 1.3, Muse Spark 1.2 | Standard PAYG ($20 buys ~10M list-blend / ~14M cached-agent tokens) Floor: 10M |
Mid: 12M |
| Contributor | Enterprise | RPM: 60 requests/min TPM: 30,000 tokens/min price: $0.10-$0.20/M contributor pricing (~95% discount via muse-spark-1.2-contributor) |
Muse Spark 1.2 (contributor) | Contributor PAYG ($20 buys ~100M conservative / ~200M optimistic tokens) Floor: 100M |
Mid: 133M |
Key Gotchas & Constraints:
- Contributor allows training on your data
- Contributor tier throttled to 60 RPM
MiniMax Token Plan (minimax)
- Primary Pricing & Docs: https://platform.minimax.io/docs/guides/pricing-token-plan
- Category:
api-provider - Data Privacy & Training: Not published on pricing page
- IP Indemnity: False
- Stacking Policy:
silent— "App terms: "your account is personal to you... not to provide any other person with access"; platform terms: "the Account should be used solely by you". Multiple paid accounts are not addressed."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Plus | $22/mo | quota: 5-hour rolling + weekly windows agents: 3-4 concurrent agents |
MiniMax M3, M2.7 | 5h + weekly windows (~20M tokens/mo est.) Floor: 20M |
RESEARCH (low) |
| Max | $55/mo | quota: 5-hour rolling + weekly windows agents: 4-5 concurrent agents |
MiniMax M3, M2.7 | Higher windows (~65M tokens/mo est.) Floor: 65M |
RESEARCH (low) |
| Ultra | $132/mo | quota: 5-hour rolling + weekly windows, extended sessions agents: 6-7 concurrent agents |
MiniMax M3, M2.7 | Highest windows (~160M tokens/mo est.) Floor: 160M |
RESEARCH (low) |
| Prepaid Credits | Enterprise | credits: $5 = 5,000 / $25 = 25,000 / $100 = 100,000 credits (1,000 credits = $1) validity: 365 days |
All MiniMax models | $1 per 1,000 credits Floor: 0M |
RESEARCH (low) |
Key Gotchas & Constraints:
- Sold as 'Token Plan' subscription rather than a separate coding plan
- Exact credits inside 5h/weekly windows are not officially published - token estimates are research-derived
- MiniMax H3, voice design and rapid voice cloning not included
- Prepaid credits: 1,000 credits = $1, valid 365 days
Mistral API (mistral-api)
- Primary Pricing & Docs: https://mistral.ai/pricing
- Category:
api-provider - Data Privacy & Training: Opt-out available
- IP Indemnity: False
- Stacking Policy:
unknown
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Free (Le Plan Studio) | Free ($0) | credits: Signup credits + $10/mo included credits | Mistral Medium 3.5, Codestral | $10/mo credits (~10M tokens on select models) Floor: 10M |
RESEARCH (low) |
| PAYG | Enterprise | batchDiscount: 50% off Batch API cacheDiscount: 90% off prefix caching hardContextCap: Enforced context limit (400 rather than truncation) |
Mistral Medium 3.5, Mistral Small 4, Mistral Large 3 (+3 more) |
Direct PAYG ($20 buys ~20M Codestral tokens) Floor: 20M |
RESEARCH (low) |
Key Gotchas & Constraints:
- Model lifecycle folded Magistral into Mistral Medium 3.5; Small is now 'Mistral Small 4'
- Training is opt-out rather than off by default; verify your workspace privacy settings
- Exceeding the context window triggers hard 400 Bad Request rather than truncation
- Codestral license restricts certain commercial reuse outside paid API
Ollama Cloud (ollama-cloud)
- Primary Pricing & Docs: https://ollama.com
- Category:
api-provider - Data Privacy & Training: No
- IP Indemnity: False
- Stacking Policy:
prohibited— "Pricing FAQ: "Can I have multiple Ollama accounts? No. Ollama is one account per person.""
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Free | Free ($0) | cloudModels: Community hosted models concurrency: 1 stream localUse: Unlimited local execution |
Llama 3.2, Qwen 2.5 Coder 7B | Free cloud tier (~1M tokens/mo) Floor: 1M |
RESEARCH (low) |
| Pro | $20/mo ($200/yr) |
credits: $60 cloud compute credits/mo concurrency: 3 concurrent streams peakSurcharge: 12:00-18:00 UTC peak rates, Mon-Fri |
DeepSeek V4.1 Flash, DeepSeek V4-Pro, MiniMax M3, GLM-5, Kimi K3 | $60 credits (~60M baseline; 31.4M Kimi K3 to 1.23B DeepSeek Flash) Floor: 60M |
RESEARCH (low) |
| Max | $100/mo | credits: $300 cloud compute credits/mo concurrency: 10 concurrent streams peakSurcharge: Standard rates |
DeepSeek V4.1 Flash, DeepSeek V4-Pro, MiniMax M3, GLM-5, Kimi K3 | $300 credits (~300M baseline; 157.01M Kimi K3 to 6.15B DeepSeek Flash) Floor: 300M |
RESEARCH (low) |
| Team | $500/mo | credits: $1,000 cloud compute credits/mo concurrency: 10 concurrent streams privateEndpoints: Dedicated pods |
DeepSeek V4.1 Flash, DeepSeek V4-Pro, MiniMax M3, GLM-5, Kimi K3 | $1,000 credits (~1B baseline; 523.36M Kimi K3 to 20.49B DeepSeek Flash) Floor: 1000M |
RESEARCH (low) |
Key Gotchas & Constraints:
- Dollar credit pools ($60 Pro, $300 Max, $1,000 Team) spent across DeepSeek V4.1 Flash, DeepSeek V4-Pro, MiniMax M3, GLM-5, and Kimi K3
- Unused included usage does not roll over
- Peak pricing applies 12:00-18:00 UTC Monday-Friday
- Stacking prohibited: one account per person is strictly enforced; multiple accounts banned, clamped to 1 copy in Dangerous Dave mode
OpenAI API (openai-api)
- Primary Pricing & Docs: https://platform.openai.com
- Category:
api-provider - Data Privacy & Training: No (API data not used for training by default; 30-day default retention, ZDR on request)
- IP Indemnity: True
- Stacking Policy:
silent— "Terms ban sharing account credentials; multiple paid accounts are not addressed."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| PAYG | Enterprise | spendTiers: $100 / $500 / $1,000 / $5,000 / $200,000 mo spend caps (official) retention: 30-day default retention; ZDR available on request for eligible endpoints batchDiscount: 50% off Batch API |
GPT-6 Astra, GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna | Direct API PAYG ($20 buys ~10M mixed-frontier / ~18M Sol-class cached-agent / ~100M Luna-class) Floor: 10M |
Mid: 18M |
Key Gotchas & Constraints:
- Long-context pricing doubles on requests exceeding 272K tokens in context window
- Rate limits (TPM/RPM) are strictly tiered based on total historical account spend
- Reasoning tokens count towards output token billing
Together.ai (together-ai)
- Primary Pricing & Docs: https://www.together.ai/pricing
- Category:
api-provider - Data Privacy & Training: No
- IP Indemnity: False
- Stacking Policy:
silent— "Terms of Service reviewed; no multiple-account, sharing or seat clause found."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Free | Free ($0) | rateLimits: Dynamic rate limits - no fixed per-model RPM/TPM published; scales with account history creditTrial: $5 free credits |
Llama 3.3 70B, DeepSeek V3, Qwen 2.5 Coder | Signup credit (~5M tokens; amount not listed on the official pricing page) Floor: 5M |
RESEARCH (low) |
| PAYG | Enterprise | tierLimits: Dynamic limits scale with spend history ($25/$50/$100/$250 milestones historically) batchDiscount: 50% off Batch inference |
Llama 3.3 70B, DeepSeek V3, Qwen 2.5 Coder 32B | PAYG ($20 buys ~30M tokens) Floor: 30M |
RESEARCH (low) |
Key Gotchas & Constraints:
- Sudden traffic spikes hit 429 rate limits until account tier auto-upgrades
- Cold start delays on infrequently accessed open-weights models
- Dedicated endpoints incur minimum hourly billing
Xiaomi MiMo Token Plan (xiaomi-mimo)
- Primary Pricing & Docs: https://mimo.mi.com/docs/en-US/price/token-plan · Token Plan Subscription FAQ: https://mimo.mi.com/docs/en-US/quick-start/faq/token-plan · Pay-As-You-Go Pricing: https://mimo.mi.com/docs/en-US/price/pay-as-you-go · Rate Limits: https://mimo.mi.com/docs/en-US/api/guidance/rate-limit · Platform: https://platform.xiaomimimo.com/token-plan
- Category:
api-provider - Data Privacy & Training: Standard API terms; no explicit opt-out mechanism documented for model training
- IP Indemnity: False
- Stacking Policy:
prohibited— "Each account can only subscribe to one Token Plan at a time. If you want to change your plan, you must wait until the current billing cycle ends."
Credit-to-Token Conversion Math
Xiaomi Token Plan uses a credit-based system where different operations consume different credit amounts per token:
| Model | Cache Hit (credits/token) | Cache Miss (credits/token) | Output (credits/token) |
|---|---|---|---|
| MiMo-V2.5 | 2 | 100 | 200 |
| MiMo-V2.5-Pro | 2.5 | 300 | 600 |
Under Token-Max's standard agentic request (20K input, 75% cache hit, 1K output):
- MiMo-V2.5: (15,000 × 2) + (5,000 × 100) + (1,000 × 200) = 730,000 credits/turn → 34.762M credits per 1M tokens
- MiMo-V2.5-Pro: (15,000 × 2.5) + (5,000 × 300) + (1,000 × 600) = 2,137,500 credits/turn → 101.786M credits per 1M tokens
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Lite | $6/mo | credits: 4,100,000,000 (4.1B) Credits/mo offPeakMultiplier: 0.8× deduction rate UTC 16:00–24:00 rateLimit: 10 RPM, 200 TPM (K) |
MiMo-V2.5-Pro, MiMo-V2.5 | 4.1B credits; V2.5-Pro 40.28M / V2.5 117.93M tokens (derived from official table) Floor: 40.28M (V2.5-Pro) / 117.93M (V2.5) |
OFFICIAL (high) "Lite: 4.1 billion credits" |
| Standard | $16/mo | credits: 11,000,000,000 (11B) Credits/mo offPeakMultiplier: 0.8× deduction rate UTC 16:00–24:00 rateLimit: 10 RPM, 200 TPM (K) |
MiMo-V2.5-Pro, MiMo-V2.5 | 11B credits; V2.5-Pro 108.07M / V2.5 316.42M tokens (derived from official table) Floor: 108.07M (V2.5-Pro) / 316.42M (V2.5) |
OFFICIAL (high) "Standard: 11 billion credits" |
| Pro | $50/mo | credits: 38,000,000,000 (38B) Credits/mo offPeakMultiplier: 0.8× deduction rate UTC 16:00–24:00 rateLimit: 10 RPM, 200 TPM (K) |
MiMo-V2.5-Pro, MiMo-V2.5 | 38B credits; V2.5-Pro 373.33M / V2.5 1093.08M tokens (derived from official table) Floor: 373.33M (V2.5-Pro) / 1093.08M (V2.5) |
OFFICIAL (high) "Pro: 38 billion credits" |
| Max | $100/mo | credits: 82,000,000,000 (82B) Credits/mo offPeakMultiplier: 0.8× deduction rate UTC 16:00–24:00 rateLimit: 10 RPM, 200 TPM (K) |
MiMo-V2.5-Pro, MiMo-V2.5 | 82B credits; V2.5-Pro 805.61M / V2.5 2358.9M tokens (derived from official table) Floor: 805.61M (V2.5-Pro) / 2358.9M (V2.5) |
OFFICIAL (high) "Max: 82 billion credits" |
Key Gotchas & Constraints:
- Credit system: costs vary by cache hit/miss/output — actual token yield depends heavily on cache hit rate
- Off-peak 0.8× multiplier (UTC 16:00–24:00) effectively gives 25% more tokens during off-peak hours
- Each account limited to one active Token Plan at a time; no mid-cycle upgrades
- MiMo-V2.5-Pro-UltraSpeed (MXFP4 quantized, >1000 TPS) is PAYG-only and NOT included in Token Plans
- Rate limits are relatively low (10 RPM, 200K TPM) compared to Western providers
- Token Plan credits do NOT roll over to the next billing cycle
- Overseas (non-China) PAYG pricing is higher than domestic pricing
- OpenRouter pricing may differ from direct API pricing
Z.ai GLM Coding Plan (z-ai)
- Primary Pricing & Docs: https://z.ai/subscribe · Usage Instructions: https://docs.z.ai/devpack/overview#usage-instruction · Usage Policy: https://docs.z.ai/devpack/usage-policy
- Category:
api-provider - Data Privacy & Training: Team plans: conversation data not used for model training (official). Personal plans: standard subscription terms, no public no-training commitment
- IP Indemnity: False
- Stacking Policy:
prohibited— "Subscription terms: benefits are "exclusive to the subscriber"; "Account sharing or multi-user access is prohibited" and the licence is tied to a single natural person with no aggregation/proxying."
Stated Limits & Token Yield Estimates
| Tier | Price | Stated Quotas & Limits | Supported Models | Monthly Token Budget | Source & Evidence |
|---|---|---|---|---|---|
| Lite | $18/mo ($151/yr) |
fiveHourCredits: 2,000 credits weeklyCredits: 10,000 credits weeklyLimitGLM53: Up to 97M Tokens/week weeklyLimitGLM53Flash: Up to 584M Tokens/week weeklyCeiling: Up to 97M GLM-5.3 / 584M GLM-5.3-Flash tokens/week (~420M / 2,529M monthly ceiling) concurrency: Recommended for 1 project at a time (dynamic off-peak boost) offPeakDiscount: 50% credit rate during off-peak peakHours: Mon-Fri 14:00-18:00 UTC+8 toolRestrictions: Strictly limited to supported coding tools (Claude Code, Cline, OpenCode, Goose) |
GLM-5.3, GLM-5.3-Flash | 10K weekly credits; GLM-5.3 48-97M tokens/wk at 95% cache (official) Floor: 208M |
Mid: 314M |
| Pro | $80/mo ($605/yr) |
fiveHourCredits: 12,000 credits weeklyCredits: 60,000 credits weeklyLimitGLM53: Up to 582M Tokens/week weeklyLimitGLM53Flash: Up to 3,504M Tokens/week weeklyCeiling: Up to 582M GLM-5.3 / 3,504M GLM-5.3-Flash tokens/week (~2,520M / 15,172M monthly ceiling; 6× Lite) concurrency: Recommended for 1-2 projects simultaneously (dynamic off-peak boost) offPeakDiscount: 50% credit rate during off-peak includedMcps: Vision, Web Search, Web Reader, Zread |
GLM-5.3, GLM-5.3-Flash | 60K weekly credits; GLM-5.3 290-582M tokens/wk at 95% cache (official) Floor: 1256M |
Mid: 1888M |
| Max | $168/mo ($1344/yr) |
fiveHourCredits: 28,000 credits weeklyCredits: 140,000 credits weeklyLimitGLM53: Up to 1,358M Tokens/week weeklyLimitGLM53Flash: Up to 8,176M Tokens/week weeklyCeiling: Up to 1,358M GLM-5.3 / 8,176M GLM-5.3-Flash tokens/week (~5,884M / 35,402M monthly ceiling; 14× Lite) concurrency: Recommended for 2+ projects simultaneously (highest allocation, dynamic off-peak boost) offPeakDiscount: 50% credit rate during off-peak priority: Priority resource allocation during peak hours |
GLM-5.3, GLM-5.3-Flash | 140K weekly credits; GLM-5.3 676M-1.36B tokens/wk at 95% cache (official) Floor: 2927M |
Mid: 4405M |
| Team Standard | $80/mo | fiveHourCredits: 15,000 credits / seat weeklyCredits: 66,000 credits / seat minSeats: 2 seats minimum concurrency: 1-2 concurrent projects per seat dataPrivacy: Excluded from model training by default overage: On-demand overage billed at a 10% discount from model API list price billing: Centralized billing, seat management & VAT invoicing |
GLM-5.3, GLM-5.3-Flash | 66K weekly credits/seat; GLM-5.3 319-638M tokens/wk/seat (derived from official table) Floor: 1381M |
Mid: 2072M |
Key Gotchas & Constraints:
- Strictly tool-only: direct generic API scraping or unauthorized tools trigger error 1113 or risk control suspension
- Dual 5-hour rolling limits and 7-day weekly reset cycles - exhaustion requires waiting for the 5-hour refresh
- Peak hours (Mon-Fri 14:00-18:00 UTC+8) consume credits at 2x the off-peak rate
- All plans serve GLM-5.3 and GLM-5.3-Flash only; GLM-5.2/GLM-5.1 requests are auto-routed to GLM-5.3 and GLM-4.7 to GLM-5.3-Flash
- GLM-5.3-Flash campaign (Sep 3-20, 2026): unlimited usage via ZCode 23:00-09:00 SGT and doubled quota on other agents - temporary bonus on top of the published allowance
- Subscriptions auto-renew at the end of each billing cycle; cancellation must be done at least 3 days before renewal date
- Subscriptions are non-refundable once purchased; downgrading team seats mid-cycle is not supported
- Account sharing or multi-user access is prohibited; violations trigger risk control (rate limiting, freezing, ban after >3 violations)
4. Cross-Reference Index
docs/DATA_SOURCES.md: End-to-end ingestion pipeline and blended calculation methodology.docs/TOKEN_ESTIMATE_VALIDATION.md: Empirical OSINT validation for 20:1 agent blends and prompt caching.docs/VERIFICATION.md: Historical verification audit log and provider test passes.data/estimate-constants.json: Canonical mathematical constants shared across frontend and data scripts.