Limits last verified Jul 29, 2026

AI coding assistant rate limits

How Cursor, GitHub Copilot, and Claude Code actually meter usage — request allowances, credits, completions, and what happens when you hit the ceiling. Each figure dated and sourced.

Every figure below is dated and linked to its official source. Unpublished caps are marked "Not published" — we never guess. Full cross-provider table: AI usage limits comparison.

Tier-by-tier limits

Side-by-side tracked limits for each tier.

LimitCursor FreeCursor ProCopilot FreeCopilot ProClaude Pro
PriceFree$20/monthFree$10/month$20/month
Messages / requests per windowNot published1,500 AI Credits per month~45 messages per 5 hours
Weekly capApplies across all models
Context window200K tokens
Model accessLimited agent requests and tab completionsAll supported frontier coding modelsBase model; limited premium-model accessUnlimited base-model chat; frontier models consume AI Credits
Agent / tool useNot publishedSet monthly model-usage allowance2,000 per monthUnlimitedIncluded (share the same usage limits)
When you hit the limitContinue via on-demand usage, billed in arrearsHard stop until the 5-hour window resets; separate weekly cap resets at a fixed weekly time

✓ Figures verified between Jan 15, 2026 and Jul 29, 2026 against official provider pages · hover any value for its date · full per-fact dates and sources in the plan sections

Cursor Hobby / Free: full tracked limits

Free · Official pricing page · Full plan guide

LimitValueVerifiedSource
Model accessLimited agent requests and tab completionsJul 29, 2026Official page
Agent requestsNot published
Hobby lists 'limited Agent requests' and Composer access without numbers (an earlier published figure was 2,000 completions/month).
Jul 29, 2026Official page

Cursor Pro: full tracked limits

$20/month · Official pricing page · Full plan guide

LimitValueVerifiedSource
Model accessAll supported frontier coding modelsJul 29, 2026Official page
Included model usageSet monthly model-usage allowance
Cursor: 'Every plan includes a set amount of model usage'; community estimates put Pro's allowance around $20 of API-rate usage. Pro+ (3× Pro) and Ultra (20× Pro) tiers extend it. Tab completions are unlimited.
Jul 29, 2026Official page
At the limitContinue via on-demand usage, billed in arrearsJul 29, 2026Official page

GitHub Copilot Free: full tracked limits

Free · Official pricing page · Full plan guide

LimitValueVerifiedSource
Chat / agent usageNot published
GitHub now lists 'limited chat and agent usage' without a number (previously 50 premium requests/month before the June 2026 AI Credits change).
Jul 29, 2026Official page
Model accessBase model; limited premium-model accessJan 15, 2026Official page
Code completions2,000 per monthJul 29, 2026Official page

GitHub Copilot Pro: full tracked limits

$10/month · Official pricing page · Full plan guide

LimitValueVerifiedSource
AI Credits (premium usage)1,500 AI Credits per month
≈$15 of model usage (1 credit = $0.01), metered by token consumption at each model's rate. Replaced the legacy 300 premium requests/month on June 1, 2026; extra credits purchasable.
Jul 29, 2026Official page
Model accessUnlimited base-model chat; frontier models consume AI CreditsJul 29, 2026Official page
Code completionsUnlimitedJul 29, 2026Official page

Claude Pro: full tracked limits

$20/month · Official pricing page · Full plan guide

LimitValueVerifiedSource
Messages~45 messages per 5 hours
Anthropic's guidance: expect at least 45 messages every 5 hours for shorter conversations — varies with message length, attachments, model, and demand.
Jul 29, 2026Official page
Weekly usage capApplies across all models
Added August 2025 on top of 5-hour windows; exact numbers are not published. Resets at a fixed time each week (shown in Settings → Usage). Mainly felt by heavy Claude Code use.
Jul 29, 2026Official page
Context window200K tokensJul 29, 2026Official page
Claude Code & CoworkIncluded (share the same usage limits)Jul 29, 2026Official page
At the limitHard stop until the 5-hour window resets; separate weekly cap resets at a fixed weekly timeJul 29, 2026Official page

Recorded limit changes

From the full changelog.

  1. Copilot Pro

    GitHub replaced Copilot's monthly premium requests with GitHub AI Credits: Pro now includes 1,500 credits (≈$15 of model usage) per month, metered by token consumption at each model's rate.

    Before: 300 premium requests / month ($0.04 per extra). After: 1,500 AI Credits / month (1 credit = $0.01).

    Source
  2. Claude Pro

    Anthropic added weekly usage caps for Claude Pro and Max subscribers on top of the existing 5-hour windows, mainly affecting heavy Claude Code use.

    Before: 5-hour windows only. After: 5-hour windows + weekly cap (~40–80 h Sonnet-class / week on Pro).

    Source
  3. Cursor Pro

    Cursor moved Pro from a fixed 500 fast requests per month to roughly $20 of included frontier-model usage with unlimited tab completions (announced mid-June 2025).

    Before: 500 fast requests / month. After: ≈$20 included model usage / month.

    Source

Practical guidance

How these limits play out in real workflows.

Three different metering models

Copilot meters by monthly AI Credits (1,500 on Pro, ≈$15 of token usage), Cursor by an included model-usage allowance with on-demand billing after, and Claude Code by Claude's 5-hour windows plus an all-model weekly cap. The same workload can feel cheap on one and throttled on another.

Agentic work burns allowances fastest

Long agent sessions consume tokens far faster than completions or chat. Copilot's unlimited completions cost nothing extra, but agent runs draw down AI Credits; on Cursor, agent requests are exactly what the included allowance meters; heavy Claude Code agent use is what Claude's weekly cap exists for.

Match the meter to your workflow

Completions-first coders get the most per dollar from Copilot Pro at $10. Agent-heavy refactoring fits Cursor Pro's usage allowance (with Pro+ at 3× and Ultra at 20× as escalation paths). If you also want a general assistant, Claude Pro bundles Claude Code into the same $20 subscription.

FAQ

Answers based on the tracked figures above.

Which paid coding AI plan has the highest usage limits?

They meter differently: Copilot Pro gives unlimited completions plus 1,500 AI Credits/month for frontier models; Cursor Pro gives a set monthly model-usage allowance (roughly $20 at API rates by community estimates) then on-demand billing; Claude Pro gives ~45 messages per 5 hours shared between chat and Claude Code. For pure completions volume, Copilot wins; for agent work, Cursor's paid allowance usually goes furthest.

What are Copilot's limits after the June 2026 change?

On June 1, 2026 GitHub replaced the 300 monthly premium requests with AI Credits: Copilot Pro includes 1,500 credits (1 credit = $0.01, ≈$15 of usage) metered by token consumption at each model's rate, with extra credits purchasable. Completions and base-model chat remain unlimited.

Does Claude Code have its own separate limit?

No — Claude Code shares your Claude plan's 5-hour windows and weekly cap. Heavy agent sessions are the main way users actually hit the weekly cap Anthropic added in August 2025.

Related pages