“Rate limited” or 429 errors from your AI
What a 429 means, how long to wait, and how to make free limits last longer.
Updated November 8, 2026
A 429 or "rate limited" message means your AI provider is refusing more requests for now. Your game and project are fine. Nothing was lost.
Which limit did I hit?
| Provider | Limit | When it resets |
|---|---|---|
| OpenRouter free models | 20 per minute, and 50 per day on a zero balance (1,000 after a one-time $10 purchase) | The per-minute limit clears within a minute. The daily one resets the next day. |
| Codex on a free ChatGPT plan | A rolling 5-hour window, plus a weekly cap | The 5-hour window refills on its own. |
| Antigravity CLI | Weekly rate limits on a rolling window | Gradually, not at midnight. |
| Google AI Studio key | 5 to 15 per minute, 100 to 1,000 per day, depending on the model | Per minute, then daily. |
What to do now
- Wait a minute and try once more. A per-minute limit clears quickly.
- Switch to another free model. On OpenRouter, each free model is a separate endpoint, and one being busy says nothing about the others. The model guide has tiers to fall back to.
- Switch AI. Run Bricklayer: Choose AI Agent (Ctrl+I) and pick a different provider for the rest of the day.
A Gemini key from Google AI Studio that returns 429 straight away, even on the first request, is a known issue with the retired Gemini CLI. Use Antigravity CLI instead. See Build without paying for AI.
Make free limits last longer
Agents often make several requests for one message: reading files, planning, editing, checking. You can cut that down:
- Ask for one feature at a time. "Add a checkpoint system" uses far fewer requests than "make the whole obby."
- Name the script. "In the CoinService script…" saves the AI from searching the whole project.
- Ask for a plan first, and approve it, so the AI does not try three approaches.
More in Write prompts that get good results.
Did this answer your question?
Related articles
Still need help? or start a thread in the community forum.