Is OpenRouter down?
The live status from status.openrouter.ai is below. If every model fails or times out, OpenRouter itself is down; a 502 or 503 on one model usually means that model's providers failed. Retry with backoff, pick another model, or use the provider's own API key.
Live OpenRouter status
Checked
All systems operational
- Chat (/api/v1/chat/completions)Operational
- Embeddings (/api/v1/embeddings)Operational
- Web & Application ServicesOperational
- Last incident: Usage Reporting Delays in Web Appresolved Oct 9
Source: status.openrouter.ai. Read on our server about every 2 minutes.
Is it OpenRouter, or one model?
| What you see | What it means |
|---|---|
| Every model fails or times out | OpenRouter's API. Check the Chat row above |
| HTTP 502 | "Your chosen model is down or we received an invalid response from it": that model's provider failed. Try another model, or let OpenRouter fall back |
| HTTP 503 | "There is no available model provider that meets your routing requirements": your provider filters (order, only, data policy) left no provider. Loosen them or pick another model |
| HTTP 429 | Rate limited, by OpenRouter or by the model's provider. See the OpenRouter 429 fix |
| HTTP 402 | Your credits ran out |
| A stream ends with an error | The status stays 200 because OpenRouter had already sent it when every provider failed; the last error is in the stream |
Five minutes into an OpenRouter problem
- If only one model fails, keep provider fallbacks on (the default) and add a
modelsarray so OpenRouter can move to a second model. - Retry with exponential backoff and honor Retry-After on 429 and 503.
- If OpenRouter itself is down, switch your agent to a direct key from the model's own provider, such as Anthropic, OpenAI or Google, until it recovers.
How to switch to a direct provider key
| Agent | Steps |
|---|---|
| Claude Code | Unset ANTHROPIC_BASE_URL and sign in with /login, or use an Anthropic API key |
| Cline, Roo Code, Kilo Code | Settings, API Provider: choose Anthropic, OpenAI or Google with that provider's key |
| OpenCode | Run opencode auth login for another provider, then /models |
| Aider | Restart with the provider's key in its environment variable and --model set to that provider's model |
Where Standard Compute stands
- Standard Compute sends model requests through OpenRouter, so when OpenRouter's API is down, Standard Compute's model requests fail too: after two retries they return HTTP 503.
- On top of OpenRouter it handles the smaller failures: a reviewed provider order for each model, two retries on 429 and 5xx, and specific model IDs that always stay that model.
The catch
For a full OpenRouter outage, keep one direct key with a model's own provider. That is the backup, for Standard Compute users too.
Where Standard Compute fits
Standard Compute is an LLM gateway for coding agents built on OpenRouter, with monthly plans from $19 (USD) and a fixed compute budget in place of per-token credits, a reviewed provider order per model and retries on 429 and 5xx. One key works in Claude Code, Codex CLI, OpenCode, Cline and Aider.
FAQ
Is OpenRouter down or is it just one model?
If every model fails or times out, check status.openrouter.ai. If only one model fails with 502 or 503, that model's providers are the problem: switch model, or keep fallbacks on and pass a models array.
Does OpenRouter fall back to another provider automatically?
Yes, by default. If a provider fails, OpenRouter tries another provider of the same model unless you set allow_fallbacks to false or restrict providers with only. To fall back to a different model, pass a models array.
Does Standard Compute keep working when OpenRouter is down?
No. Standard Compute sends its model requests through OpenRouter, so an OpenRouter API outage reaches it too; requests return HTTP 503 after two retries. Keep a direct provider key as the backup for that case.
What does 'There is no available model provider that meets your routing requirements' mean?
OpenRouter found no provider that matches the model and your provider settings, such as order, only, ignore or a data policy. It is usually a settings problem: loosen the provider filters or pick another model.