Is Cohere down?
Cohere
No — Cohere is working normally
Our sources see nothing wrong. If it still fails for you, the cause is most likely on your side — the tips further down cover the usual suspects.
What our sources say
- Official status page
- Working
- Our test requests, last hour
- Not running for this service
- Traffic through OpenRouter
- 0 of 3 models failing, 100% of requests succeeded in 30 min
- Reports from visitors, last hour
- 0
Parts of the service
- All Endpoints Working
- embeddings Working
- All Models Working
- command-a-plus-05-2026 Working
- command-a-reasoning-08-2025 Working
- command-a-vision-07-2025 Working
- command-a-translate-08-2025 Working
- command-a-03-2025 Working
- command-r-plus-08-2024 Working
- command-r-08-2024 Working
- command-r7b-12-2024 Working
- embed-v4.0 Working
- embed-english-v3.0 Working
- embed-english-light-v3.0 Working
- embed-multilingual-v3.0 Working
- embed-multilingual-light-v3.0 Working
- rerank-v4.0-fast Working
- rerank-v4.0-pro Working
- rerank-v3.5 Working
- rerank-english-v3.0 Working
- rerank-multilingual-v3.0 Working
- c4ai-aya-vision-32b Working
- c4ai-aya-expanse-32b Working
- tiny-aya-base Working
- tiny-aya-global Working
Last 24 hours
24 h agonow
Last 30 days
30 days agoyesterday
30 of 30 days with data had no problems.
Recent incidents
- Slow or glitchy
Resolved after 4 hrs 22 mins
- Partly down
Documentation Content Loading Issues
Resolved after 57 mins
About this service
Cohere's API covers chat (Command), embeddings and rerank. An embeddings or rerank problem can quietly break search inside apps built on Cohere even while chat answers normally.
Other model APIs right now
Cohere: common questions
Is the Cohere API down, or is it my code?
Look at the HTTP status. 401 and 403 mean a key or permission problem, 400 and 422 a malformed request, 429 a rate or spending limit. Only 500, 502, 503, 504 and timeouts point at Cohere. If this page is green while you get them, send a minimal request with curl to rule out your client library or proxy.
How should my app handle Cohere outages?
Retry 429 and 5xx responses with exponential backoff and a little random jitter, cap the number of attempts, and set a timeout so stuck streams don't pile up. For anything users see, keep a fallback model from another vendor configured and switch to it automatically after a few failures in a row — the list above shows which ones are healthy right now.
Why do my requests fail when others say it works?
Limits apply per organisation and per model, so your account can be throttled while everyone else is fine. Newly released and very large models are also the first to be rate-limited at busy hours. Check the usage page of your account, try a smaller model and make sure you call the right base URL and region.
Responses are slow but they arrive. Does that count as an outage?
Watch the time to the first token rather than the total time. On our own test requests we flag a slowdown when the first token takes more than 3 seconds and three times longer than the day's median. Long prompts and long outputs are slow by nature and are not treated as a problem.
How do you know whether Cohere is down?
For Cohere we combine Cohere's official status page, read every minute, the share of successful requests to Cohere's models on OpenRouter and reports from people who press the button on this page. The most serious signal that is still fresh sets the verdict, and stale sources are ignored rather than trusted. The page refreshes by itself; the exact thresholds are on our “How we check” page.