Hit my AI coding usage limit. What now?

Updated 2026-10-17 · 4 min read · limits and prices from each vendor's own pages

TL;DR: 1) Check the reset time (Claude Code and Codex use a rolling 5-hour window plus a weekly cap). 2) Switch to a lighter model. 3) Turn on pay-as-you-go if your plan offers it. 4) Or switch to a backup from a different company, which has its own separate limits: GitHub Copilot, Devin Desktop (formerly Windsurf), Google Antigravity, Cline, Kiro all have free tiers.
ToolHow the limit works (checked 2026-10-17)When you hit it
Claude Code
vendor estimates
Rolling 5-hour session limit + weekly limit, shared with Claude chat. Pro ≈ 10–40 Claude Code prompts per 5h (older Anthropic estimate); Max 5x/20x more per session. source (opens in a new tab)Wait for the 5h/weekly reset, turn on usage credits (API rates), or upgrade to Max.
OpenAI Codex
vendor estimates
5-hour window + weekly limit; local messages and cloud tasks share it. Plus ≈ 15–160 local messages/5h on mid-tier models, far fewer on the top model (OpenAI estimates; tables vary by model). Limited-time: Codex also on ChatGPT Free/Go. source (opens in a new tab)Switch to a lighter model, buy ChatGPT credits or a paid instant reset (eligible Plus/Pro), or use an API key.
Cursor
no numbers published
Usage pool shared with Grok that refreshes weekly: Plus = base (Grok models only), Premium 4x, Super 15x, Ultra 50x Plus. No per-message numbers published. source (opens in a new tab)Buy top-offs, upgrade, or wait for the weekly refresh. On Premium and above, third-party models are billed separately at API list price.
Devin Desktop (formerly Windsurf)
vendor estimates
Daily + weekly quota (token-based; free models don't count). Pro: ~7–27 premium-plus, 8–101 premium or 47–190 lightweight messages/day (Mar 2026 estimate); Max: bigger weekly quota, no daily cap. source (opens in a new tab)Buy extra usage at API pricing, use lightweight models, or wait for daily/weekly reset.
Kiro
published numbers
Monthly credits, used fractionally per request: Free 50 (rate-limited, weekly quota), Pro 1,000, Pro+ 2,000, Pro Max 5,000, Power 10,000. source (opens in a new tab)Paid tiers buy add-on credits at $0.04 each; Free waits for reset.
GitHub Copilot
published numbers
Free: 2,000 completions/mo + limited chat/agent. Pro: 1,500 AI credits/mo · Pro+: 7,000 · Max: 20,000. source (opens in a new tab)Overage is pay-as-you-go at $0.01 per AI credit, or upgrade.
Google Antigravity
no numbers published
Free: 'basic weekly rate limits'; paid tiers get more plus a flexible AI credit pool. No numbers published. source (opens in a new tab)Wait for the weekly refresh or upgrade.
Cline
published numbers
No plan limits: you pay your model provider per token (or use Cline's inference). Limits are your budget and provider rate limits. source (opens in a new tab)Top up your API account or switch to a cheaper/free model.
Lovable
published numbers
Free: 5 credits/day, max 30/mo. Pro: 100 credits/mo + 5/day. Plan mode ≈ 1 credit/message; Build mode is usage-based. source (opens in a new tab)Buy top-up credits or upgrade.
Bolt.new
published numbers
Free: 300K tokens/day, 1M/mo. Pro: no daily limit, from 10M tokens/mo; unused tokens roll over 1 month. source (opens in a new tab)Buy token reloads or a bigger Pro tier.
Replit Agent
no numbers published
Free Mode Agent work within each plan's allowance: Starter has a daily allowance; Core/Pro reset every 5 hours with weekly limits. Paid modes are effort-based. No exact numbers published. source (opens in a new tab)Pay for extra usage or wait for reset.
Rosebud AI
published numbers
Free: 8,000 credits refilled weekly (≈4 edits with the default AI at ~2,000 credits each). Paid: daily refresh credits + monthly credits (300K on 10x Dev, 1M on Pro). source (opens in a new tab)Buy a credit top-up or upgrade; free credits refill weekly.

Why do I keep hitting the weekly limit?

Most agents now have two limits: a short window (5 hours for Claude Code and Codex, daily for Devin Desktop) and a weekly cap. Long agent runs, big repos, parallel sessions and top-tier models burn through both fast. Using a lighter model for routine edits stretches the week.

What's the fastest way to keep working right now?

Claude: turn on usage credits (billed at API rates) or move to Max. Codex: switch to a lighter model, buy ChatGPT credits, or buy a paid instant reset (eligible Plus/Pro). Devin Desktop (formerly Windsurf), Kiro and Copilot: buy extra usage. Cursor: upgrade your tier. Cline: just top up your API key.

Which backup agent should I use?

Pick one from a different company so its limits are separate: Google Antigravity (free, weekly limits), GitHub Copilot Free (2,000 completions/month), Devin Desktop Free (formerly Windsurf, light quota), Kiro Free (50 credits) or Cline with your own API key. If your main is Claude Code, Codex or Cursor makes a good paid backup, and the other way round.

Should I just upgrade to the $100–200 plan?

Only if you hit limits most days. A $20 main plus a free backup costs less and covers most limit days. Upgrade when the backup's quality isn't good enough for your work.

Not sure? Select your class in 3 questions.

Get a main pick plus a backup for limit days.

Select your class
Patch Notes · weekly

Patch Notes: what shipped in AI agents this week

New coding agents, model drops, and every price or limit change, in a five-minute read. No spam.

  • Bolt Lite ($9) closed to new sign-ups; Pro stays $25.
  • Cursor confirmed its referral programme is discontinued.
  • Gemini CLI stopped serving free and Google AI Pro/Ultra users; replaced by Antigravity CLI.

Coming soon

The weekly "What shipped in AI agents" digest launches soon.