What the error looks like
A typical free-tier error, as posted on Google's developer forum, has these parts:
Example 429 from the free tier
429 RESOURCE_EXHAUSTED
You exceeded your current quota, please check your plan and billing details.
Quota exceeded for metric: generativelanguage.googleapis.com/generate_content_free_tier_requests, limit: 50
quotaId: GenerateRequestsPerDayPerProjectPerModel-FreeTier
Please retry in 11.615098148s.quotaMetrictells you what ran out.generate_content_free_tier_requestscounts requests on the free tier.quotaIdtells you the window.PerDayis the daily quota. APerMinuteid is the minute limit.limitis your quota for that model in that window.Please retry ingives the wait time that Google suggests.
The three 429 codes and one 403
Google's API errors page lists three error codes with HTTP 429, and a related 403:
| Code | HTTP | Meaning | Fix |
|---|---|---|---|
rate_limit_exceeded | 429 | Per-minute or per-second request or token limit hit | Retry with exponential backoff |
quota_exceeded | 429 | Daily quota used up | Wait for the reset, or request an increase |
too_many_requests | 429 | Too many requests in a short window | Retry with exponential backoff |
permission_denied | 403 | The project is restricted, or the free tier is unavailable for it | Link a billing account and use the paid tier |
A minute limit clears in less than one minute. A daily quota clears only at the reset. Google resets requests per day at midnight Pacific time. Retrying a daily quota error before the reset wastes requests.
Free tier limits today
Google applies limits to each Google Cloud project, not to each key. Limits are different for each model. Google shows your active limits in AI Studio. The table shows the free Gemini and Gemma limits in the FreeLLMAPI catalog.
| Model | Free on | Context | Free limits (best offer) | Tools | Vision |
|---|---|---|---|---|---|
| Gemini 3.8 Flash | Google AI Studio | 1M | Recurring provider quota | Yes | Yes |
| Gemini Flash (latest alias) | Google AI Studio | - | 10 req/min 20 req/day 250,000 tokens/min ~3M | Yes | Yes |
| Gemini 3.7 Flash | Google AI Studio | 1M | 10 req/min 20 req/day 250,000 tokens/min ~3M | Yes | Yes |
| Gemini 3.6 Flash | Google AI Studio | 1M | 10 req/min 20 req/day 250,000 tokens/min ~3M | Yes | Yes |
| Gemini 3.5 Flash | Google AI Studio | 1M | 10 req/min 20 req/day 250,000 tokens/min ~3M | Yes | Yes |
| Gemini 3.8 Live | Google AI Studio | - | 10 req/min 20 req/day 250,000 tokens/min ~3M | Yes | Yes |
| Gemini 3.8 Live Extended Thinking | Google AI Studio | - | 10 req/min 20 req/day 250,000 tokens/min ~3M | Yes | Yes |
| Gemini 3.8 Flash TTS | Google AI Studio | - | Ongoing AI Studio free-tier quota; account limits apply | No | No |
| Gemini 3.8 Flash-Lite TTS | Google AI Studio | - | Ongoing AI Studio free-tier quota; account limits apply | No | No |
| Gemini Robotics-ER 2 Preview | Google AI Studio | 131K | 10 req/min 20 req/day 250,000 tokens/min ~3M | No | Yes |
| Gemini 3.5 Flash Lite | Google AI Studio | 1M | 15 req/min 20 req/day 250,000 tokens/min ~3M | Yes | Yes |
| Gemini Flash Lite (latest alias) | Google AI Studio | - | 10 req/min 20 req/day 250,000 tokens/min ~3M | Yes | Yes |
Showing 12 of 21 matching free models. See all models.
Live data from the FreeLLMAPI catalog (version 2026.11.19, checked 19 November 2026). Free limits change often. The catalog is checked every day.
Most Flash models allow about 20 requests per day on the free tier. Gemma models allow many more requests per day, but they have a smaller context window. Google says that the specified limits are not guaranteed and that actual capacity may vary.
Fixes that work
- Read the
quotaId. If it containsPerDay, stop and wait for the reset at midnight Pacific time. - For a minute limit, retry with exponential backoff and jitter. Set a maximum number of retries.
- Move simple tasks to a model with a higher daily cap, for example a Gemma or Flash-Lite model.
- Send fewer tokens. Large prompts reach the tokens per minute limit fast.
- Do not use a very small
max_tokens. Gemini Flash can use hidden thinking tokens inside the output cap.
| What you try | Does it help? | Why |
|---|---|---|
| Create a new API key in the same project | No | Limits apply to the project. Google says a new key does not resolve the error. |
| Upgrade the project to the paid tier | Gives paid quota | Google says that the paid tier does not give you free-tier quota. Paid use is billed. |
Wait for the retry in time | Yes, for minute limits | The minute window rolls over |
| Change to a different model | Yes | Each model has its own quota |
Check your quota before you send
You can see the problem before the error. In AI Studio, open the rate limit page from the Gemini API docs. It shows the active limits of your project for each model. The Projects page shows your tier.
If you use FreeLLMAPI, ask the gateway which models are ready now. A model with the status exhausted has keys that all sit on cooldown. That status clears by itself when the cooldowns end.
List models that can answer now
curl "http://localhost:3001/v1/models?execution_status=ready" \
-H "Authorization: Bearer <unified-key>"Plan your daily use with simple numbers. At 20 requests per day, one Flash model gives you about one request every 72 minutes across a full day. A coding agent can use 20 requests for one task. Put the Flash models on hard tasks only, and send simple tasks to models with larger daily caps.
Fail over to other free providers
With about 20 Flash requests per day, one Gemini key is not enough for agent work. FreeLLMAPI puts Gemini behind one endpoint with other free providers. When Gemini sends a 429, the router sends the request to the next model in your chain.
- The router counts RPM, RPD, TPM and TPD for each model and key. It stops before it reaches a known limit.
- A short 429 puts the key on a 90 second cooldown. A full daily quota uses a longer cooldown: 2 minutes, 10 minutes, 1 hour, then 24 hours.
- If Google gives a retry time that is longer, the router uses Google's value.
- The
X-Fallback-Trailheader lists each failed hop, for examplegoogle/gemini-2.5-flash key1=rate_limited.
Send Gemini traffic through the gateway
curl -fsSL https://freellmapi.co/install.sh | bash
curl -i http://localhost:3001/v1/chat/completions \
-H "Authorization: Bearer <unified-key>" \
-H "Content-Type: application/json" \
-d '{"model": "auto", "messages": [{"role": "user", "content": "hi"}]}'Gemini CLI can also use the native /v1beta surface of the gateway. The gateway does not add Gemini quota. It uses other providers when Gemini is at its limit.
Known Gemini quirks
| Provider | Known issue | What to do |
|---|---|---|
| Google AI Studio | Thinking tokens use output cap | Gemini Flash can spend hidden thinking tokens inside maxOutputTokens. Avoid tiny max_tokens values or set thinkingBudget to 0 when provider support is available. |
Live data from the FreeLLMAPI catalog (version 2026.11.19, checked 19 November 2026). Free limits change often. The catalog is checked every day.
Frequently asked questions
What does 429 RESOURCE_EXHAUSTED mean in the Gemini API?
It means that your project went over a rate limit or a quota. The quotaId in the error body tells you if it is a minute limit or a daily quota.
When does the Gemini free tier quota reset?
Requests per day reset at midnight Pacific time. Minute limits reset within one minute.
Why does a new Gemini API key not fix the quota error?
Google applies limits to the Google Cloud project, not to the key. Every key in the same project shares the same quota.
What does limit: 0 mean in a Gemini quota error?
It means that your project has no free quota for that model in that window. Some projects are not eligible for the free tier. Check your limits in AI Studio or try a different model.
How do I avoid Gemini 429 errors in a coding agent?
Use several free providers behind one gateway. When Gemini reaches its limit, the gateway sends the request to another provider, so the agent continues.
Related guides
- Free LLM API rate limits
- free gemini api
- Gemini CLI with a custom endpoint and free models
- Groq rate limit reached (429)
- NVIDIA NIM API rate limit
Sources
- Google AI Studio: Rate limits (checked 20 November 2026)
- Google AI Studio: API errors (checked 20 November 2026)
- Google AI Studio: Troubleshooting (checked 20 November 2026)
- Google developer forum: free tier quota error example (checked 20 November 2026)
- FreeLLMAPI docs: Quota and cooldown engine (checked 20 November 2026)
- FreeLLMAPI docs: Response headers (checked 20 November 2026)