Home / Guides / AI Coding Usage Limits Database
AI coding usage limits database: Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI
A bounded, row-level database of how each of 5 providers' own documentation describes its AI-coding usage limits — one row per provider, sourced to that provider's own page.
Last updated 2026-08-17 · Every row independently verified 2026-08-17 · By Soren Starck
Read this before the table: what this page is and is not
- Bounded to 5 providers. Claude Code, Codex, Cursor, GitHub Copilot, and Gemini CLI — the exact set assigned to this database. No other provider is added without a new review pass.
- Not a restatement of SessionWatcher pricing or app behavior. Every claim below is about what a named provider's own documentation states, sourced to that provider's page.
- Explicit about non-disclosure. Where a provider does not publish an exact numeric allowance, the row says so instead of inferring a number.
- Explicit about current vs. retired status. Where a provider has retired or partially retired an access path, the row says so rather than presenting every row as uniformly current.
The database
One row per provider, in the order Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI. Every unit, window, disclosure status, and route status is shown with its first-party source and verification date directly beside it.
| Provider | Unit / window | Disclosure | Route status | Source & verified |
|---|---|---|---|---|
| Claude Code | Unit: Session and weekly usage windows (no fixed message/token count published) Window: A rolling five-hour session window, plus a separately tracked weekly usage window | Not publicly disclosed Anthropic's support article names the five-hour session and weekly windows for Pro, Max, Team, and seat-based Enterprise plans, and says usage appears as progress bars under Settings > Usage, but it does not state a fixed message or token count for either window. It says usage varies with message length, file attachment size, conversation length, tool usage, model choice, effort level, and artifact creation, and points readers to per-plan pages rather than stating a number itself. | Current No retirement or suspension is documented for this mechanism as of the verification date below. | Anthropic — "Usage limit best practices" → Also: Anthropic — "How do usage and length limits work?" Verified 2026-08-17 Actively tracked; quarterly re-check, next due 2026-11-17. |
| Codex | Unit: Per-model message-count ranges within a shared session window, by plan tier Window: A rolling five-hour window shared between local Codex messages and cloud chats; the source also states additional weekly limits may apply | Range disclosed The live pricing page discloses per-model message-count ranges inside the five-hour window for each plan (wider ranges on higher tiers), rather than one fixed number per plan — usage depends on the model and task, so no single exact message count is published. | Current As verified below, the five-hour window is described as active, with no suspension mentioned on this page. This site's own, separately published Codex rate-limits guide (app/guides/codex-rate-limits-explained), last updated 2026-08-03, recorded that OpenAI suspended this same rolling window on 2026-07-12 and called the suspension temporary. That is this site's own prior editorial record, not independently re-sourced to a live OpenAI page in this pass, so it is noted here as historical context rather than as an additional first-party citation for this row. | OpenAI — Codex pricing → Redirects to https://learn.chatgpt.com/docs/pricing Verified 2026-08-17 Actively tracked; quarterly re-check, next due 2026-11-17. |
| Cursor | Unit: Two separate monthly usage pools: a Cursor Models pool and an Other Models (third-party) pool Window: Both pools reset with the monthly billing cycle | Range disclosed Cursor discloses a dollar floor for the Other Models pool per tier (Pro described as including "at least $20" of third-party model usage monthly, with higher floors named on higher tiers), not an exact cap. The Cursor Models pool is described only as "Generous included usage," with no specific number disclosed. | Current No retirement or suspension is documented for this mechanism as of the verification date below. | Cursor — Models & Pricing → Verified 2026-08-17 Actively tracked; quarterly re-check, next due 2026-11-17. |
| GitHub Copilot | Unit: GitHub AI Credits, a monthly allowance per plan Window: Monthly. A separate, older GitHub page on premium-request counters states resets occur on the 1st of the calendar month at 00:00:00 UTC, but the AI Credits page itself does not restate that reset date. | Exact numeric disclosed GitHub's individual-plans documentation discloses exact monthly AI Credit totals for paid individual plans: Pro 1,500 (1,000 base + 500 flex), Pro+ 7,000 (3,900 base + 3,100 flex), Max 20,000 (10,000 base + 10,000 flex). The Free tier discloses a code-completion count (up to 2,000/month) but not a numeric AI Credits allowance. Business/Enterprise individual allowances are not covered on this page. | Current GitHub's 2026-06-01 changelog (already logged in config/change-history.mjs) announced usage-based billing measured in AI Credits as active across all plans; this row's terms are current per that same transition and are not restated as new here. A separate GitHub page (copilot-requests) still uses "premium request" terminology for reset cadence — an unreconciled vocabulary difference on GitHub's own site that this row surfaces rather than resolves. | GitHub — Copilot individual plans (AI Credits) → Also: GitHub — Copilot requests (monthly reset cadence) Verified 2026-08-17 Actively tracked; quarterly re-check, next due 2026-11-17. |
| Gemini CLI | Unit: Access-path eligibility for Gemini CLI / Code Assist IDE extensions (not a numeric quota); the underlying Gemini API does not publish a fixed rate-limit table Window: Not applicable to the access-path change itself. The Gemini API rate-limits page states limits are viewable per-account in Google AI Studio rather than published as a static per-minute/per-day table. | Not publicly disclosed Google's rate-limits documentation states limits "depend on a variety of factors (such as your usage tier)" and directs users to view their active limits in Google AI Studio rather than publishing fixed RPM/RPD/TPM numbers for interactive use. | Partially retired Google announced that, effective 2026-06-18, Gemini CLI and the Gemini Code Assist IDE extensions stop serving Google AI Pro/Ultra subscribers and free individual Code Assist users. Standard/Enterprise organization access and paid Gemini / Gemini Enterprise Agent Platform API-key access continue unaffected, per the same announcement. This matches the record already logged in config/change-history.mjs; this row does not restate it as new, only marks the access-path status explicitly for this database. | Google — Gemini CLI transition announcement → Also: Google — Gemini API rate limits Verified 2026-08-17 Actively tracked; quarterly re-check, next due 2026-11-17. |
- Unit
- Session and weekly usage windows (no fixed message/token count published)
- Window
- A rolling five-hour session window, plus a separately tracked weekly usage window
- Disclosure detail
- Anthropic's support article names the five-hour session and weekly windows for Pro, Max, Team, and seat-based Enterprise plans, and says usage appears as progress bars under Settings > Usage, but it does not state a fixed message or token count for either window. It says usage varies with message length, file attachment size, conversation length, tool usage, model choice, effort level, and artifact creation, and points readers to per-plan pages rather than stating a number itself.
- Route status detail
- No retirement or suspension is documented for this mechanism as of the verification date below.
Source & verification
Anthropic — "Usage limit best practices" →Also: Anthropic — "How do usage and length limits work?"
Verified 2026-08-17
Actively tracked; quarterly re-check, next due 2026-11-17.
- Unit
- Per-model message-count ranges within a shared session window, by plan tier
- Window
- A rolling five-hour window shared between local Codex messages and cloud chats; the source also states additional weekly limits may apply
- Disclosure detail
- The live pricing page discloses per-model message-count ranges inside the five-hour window for each plan (wider ranges on higher tiers), rather than one fixed number per plan — usage depends on the model and task, so no single exact message count is published.
- Route status detail
- As verified below, the five-hour window is described as active, with no suspension mentioned on this page. This site's own, separately published Codex rate-limits guide (app/guides/codex-rate-limits-explained), last updated 2026-08-03, recorded that OpenAI suspended this same rolling window on 2026-07-12 and called the suspension temporary. That is this site's own prior editorial record, not independently re-sourced to a live OpenAI page in this pass, so it is noted here as historical context rather than as an additional first-party citation for this row.
Source & verification
OpenAI — Codex pricing →Redirects to https://learn.chatgpt.com/docs/pricing
Verified 2026-08-17
Actively tracked; quarterly re-check, next due 2026-11-17.
- Unit
- Two separate monthly usage pools: a Cursor Models pool and an Other Models (third-party) pool
- Window
- Both pools reset with the monthly billing cycle
- Disclosure detail
- Cursor discloses a dollar floor for the Other Models pool per tier (Pro described as including "at least $20" of third-party model usage monthly, with higher floors named on higher tiers), not an exact cap. The Cursor Models pool is described only as "Generous included usage," with no specific number disclosed.
- Route status detail
- No retirement or suspension is documented for this mechanism as of the verification date below.
Source & verification
Cursor — Models & Pricing →Verified 2026-08-17
Actively tracked; quarterly re-check, next due 2026-11-17.
- Unit
- GitHub AI Credits, a monthly allowance per plan
- Window
- Monthly. A separate, older GitHub page on premium-request counters states resets occur on the 1st of the calendar month at 00:00:00 UTC, but the AI Credits page itself does not restate that reset date.
- Disclosure detail
- GitHub's individual-plans documentation discloses exact monthly AI Credit totals for paid individual plans: Pro 1,500 (1,000 base + 500 flex), Pro+ 7,000 (3,900 base + 3,100 flex), Max 20,000 (10,000 base + 10,000 flex). The Free tier discloses a code-completion count (up to 2,000/month) but not a numeric AI Credits allowance. Business/Enterprise individual allowances are not covered on this page.
- Route status detail
- GitHub's 2026-06-01 changelog (already logged in config/change-history.mjs) announced usage-based billing measured in AI Credits as active across all plans; this row's terms are current per that same transition and are not restated as new here. A separate GitHub page (copilot-requests) still uses "premium request" terminology for reset cadence — an unreconciled vocabulary difference on GitHub's own site that this row surfaces rather than resolves.
Source & verification
GitHub — Copilot individual plans (AI Credits) →Also: GitHub — Copilot requests (monthly reset cadence)
Verified 2026-08-17
Actively tracked; quarterly re-check, next due 2026-11-17.
- Unit
- Access-path eligibility for Gemini CLI / Code Assist IDE extensions (not a numeric quota); the underlying Gemini API does not publish a fixed rate-limit table
- Window
- Not applicable to the access-path change itself. The Gemini API rate-limits page states limits are viewable per-account in Google AI Studio rather than published as a static per-minute/per-day table.
- Disclosure detail
- Google's rate-limits documentation states limits "depend on a variety of factors (such as your usage tier)" and directs users to view their active limits in Google AI Studio rather than publishing fixed RPM/RPD/TPM numbers for interactive use.
- Route status detail
- Google announced that, effective 2026-06-18, Gemini CLI and the Gemini Code Assist IDE extensions stop serving Google AI Pro/Ultra subscribers and free individual Code Assist users. Standard/Enterprise organization access and paid Gemini / Gemini Enterprise Agent Platform API-key access continue unaffected, per the same announcement. This matches the record already logged in config/change-history.mjs; this row does not restate it as new, only marks the access-path status explicitly for this database.
Source & verification
Google — Gemini CLI transition announcement →Also: Google — Gemini API rate limits
Verified 2026-08-17
Actively tracked; quarterly re-check, next due 2026-11-17.
Notes on this pass
Codex's row records this site's own, separately published Codex rate limits guide as historical context for a previously documented suspension of the shared five-hour window — that claim was not independently re-sourced to a live OpenAI page in this pass, so it is carried as historical context, not as an additional first-party citation, inside that row's route-status detail above.
Copilot's row surfaces an unreconciled vocabulary difference on GitHub's own site: the primary source describes AI Credits, while a separate GitHub page cited for reset cadence still uses "premium request" terminology. This database does not resolve that difference, only records it.
Gemini's row matches the consumer Gemini CLI access-path retirement already logged in the provider plan & limit change history; this row does not restate that as a new finding, only carries the current/partially-retired distinction into this database's route-status field.
Maintenance protocol
Quarterly re-check of every row in config/usage-limits-database.mjs, next due 2026-11-17. Each row's source is re-opened to confirm it still resolves and that the unit/window/disclosure wording it supports still matches what the source currently shows; a previously not-publicly-disclosed value that has since been published, or a route status that has since changed, is updated in that row alone. No row's disclosureStatus, routeStatus, or verifiedOn may be copied forward into a later edit of this file without that row actually being re-checked against its live source on the new date.
Providers covered
This database covers 5 of SessionWatcher's supported providers: Claude Code, Codex, Cursor, GitHub Copilot, and Gemini CLI. See the full supported-provider map for every provider SessionWatcher tracks (registry verified 2026-08-11).
SessionWatcher is an independent product and is not affiliated with, endorsed by, or sponsored by Anthropic, OpenAI, Cursor, GitHub, Microsoft, or Google. Product and provider names are trademarks of their respective owners.
Measurement note for this page
This is a Lane B/C authority-asset experiment (candidate c_00dffc262b93). Its frozen success criterion is at least 60 cumulative Google Search Console impressions for this exact URL, restricted to queries containing at least one provider token and at least one limit/quota token, over the first 56 complete days after it becomes indexable. That clock has not started: the first complete indexable day has not yet been verified.
No page-level baseline exists — this URL is new. A separate, historical, directional-only reading (341 combined visitors across two existing rate-limit guides, 2026-07-07 to 2026-07-13, per ANALYTICS_WEEKLY_ANALYSIS.md) is recorded in config/usage-limits-gate.mjs and is never compared to this page's own reading as though it were a baseline for it.
Tracked internal-link clicks from this page (guide_internal_link_clicked with source=ai-coding-usage-limits) are diagnostics, not success criteria. Full contract: config/usage-limits-gate.mjs; reviewer record: AI-CODING-USAGE-LIMITS-GATE.md.