Is the OpenRouter API reachable right now?
A live probe of openrouter.ai from NetOkay’s edge, a test from your own browser, and how to tell an OpenRouter problem from an upstream model provider or your own network.
In short: The card below asks openrouter.ai for
/api/v1/models, a public endpoint that returns 200 with the model list and
needs no token. If it answers here, OpenRouter’s API is up; a stalled model request is
then either the upstream provider or your path, and the browser test tells you which side
to look at first.
Last updated
Live probe · openrouter.ai
Where is openrouter.ai reachable from?
Loading the latest observation of openrouter.ai… Test from this browser →
From this browser
—Not run. One request from your browser, sent only when you ask.
What this proves, and what it does not
Loading…
OpenRouter specifics
- API host
-
openrouter.ai, with the API under/api/v1/. The models list is public, so this probe expects a 200 rather than a 401. - Streaming
-
OpenAI‑style server-sent events. While an upstream model is still working, OpenRouter
sends SSE comment lines such as
: OPENROUTER PROCESSINGto keep the connection alive. A client that treats comment lines as data will fail to parse them; a proxy that strips comments removes the keep‑alive. - Three parties
- A request passes through your network, OpenRouter, and the upstream provider that serves the model. OpenRouter reports upstream failures inside the response body with the provider’s status code, so a 200 stream can still carry an error object.
- Official status
- status.openrouter.ai covers OpenRouter itself; upstream provider incidents are reported on their own status pages.
- Support ticket header
-
Quote the
X-Generation-Idresponse header, or thegen-…id in the body, when reporting a failure. The OpenAI SDK’s_request_idreadsx-request-id, so it will not show this value. - Client defaults
- Most clients reuse the OpenAI SDK, so the defaults are a 10-minute timeout and 2 retries. The retries apply to OpenRouter’s response, not to an upstream error carried inside a 200 stream, which the SDK treats as a completed request.
What OpenRouter status codes mean
| Code | Sent by | Meaning |
|---|---|---|
| 200 | API application | The public models list. A model request can still carry an upstream error object inside a 200 stream. |
| 401 | API application | Invalid key. |
| 402 | API application | Insufficient credits. |
| 429 / 502 | API application | 429 is a rate limit, sometimes passed through from the upstream provider; 502 means the chosen provider failed and no fallback succeeded. |
| Every row is an HTTP answer: the host was reached and the request was refused or served. None of them means openrouter.ai is unreachable. | ||
Reachable here but failing for you?
- Run the 30‑second idle stream from the network where your app fails. OpenRouter’s comment keep‑alives only help if your path passes SSE comments through unbuffered. Run the idle stream →
- Read the response body of a hanging or failed request: an error object inside a 200 stream names the upstream provider, which is then the place to look, not your network. Fixes in the streaming guide →
- Compare environments if this browser passes but a container, server or agent fails, then run the CLI from the failing environment against openrouter.ai. Compare environments →
OpenRouter API status FAQ
My stream shows odd lines starting with a colon. Is that an error?
No. Lines beginning with a colon are SSE comments; OpenRouter uses them as keep‑alives while a provider is still generating. A correct SSE parser ignores them.
The probe is 200 but my model request hangs. What next?
The probe only proves OpenRouter’s host answers. A hang is usually the upstream provider or an idle limit on your path. Run the homepage idle stream check, then try the same model with a different provider preference.
Why is this probe 200 when other providers show 401?
OpenRouter’s models endpoint is public. Other providers require a key for every endpoint, so their healthy answer to the placeholder token is 401 from the API, or 403 when their edge refuses the request before it reaches the API.