AI tools status
Assistants and model APIs that other products now depend on. All 5 are checked every 20 minutes; the status beside each one is the result of its most recent check.
AI services degrade far more often than they fail, and a slow or truncated answer is a different event from an outage - though both are reported the same way.
How reliable ai tools has been, last 30 days
Measured by our own checks, run every 20 minutes, not taken from any provider’s status page. 5 of 5 services have enough history to report.
| Service | Now | Uptime 30d | Downtime | Response |
|---|---|---|---|---|
| ChatGPT | Up | 100.0% | none | 125 ms |
| Claude | Up | 100.0% | none | 167 ms |
| GitHub Copilot | Up | 100.0% | none | 159 ms |
| Google Gemini | Up | 100.0% | none | 297 ms |
| OpenAI API | Up | 100.0% | none | 231 ms |
ChatGPT has been the steadiest here over the last 30 days. Uptime is observed uptime: The share of measured time our checks got a usable answer. It excludes nothing, including announced maintenance.
Why ai tools services go down
Model serving is capacity-bound in a way ordinary web serving is not. Each request occupies specialised hardware for seconds rather than milliseconds, so there is no spare headroom to absorb a spike. When demand rises, these services shed load deliberately: Rate limits tighten, queues lengthen, responses get slower. The service is working exactly as designed and is unusable.
The second pattern is dependency. A growing share of ordinary software now calls a model API in the background, so an AI provider outage surfaces in products with no visible AI feature at all - a support widget, a search box, a summary that never appears. This category has quietly become infrastructure.
What an outage here looks like
Requests are accepted and take far longer than usual, or return a rate-limit error. Answers may truncate mid-sentence. The interface stays perfectly responsive, which is why this reads as a slow service rather than a broken one.
When these services are under most strain
US working hours, and sharply after any major model launch. Capacity is provisioned against a forecast, so the worst incidents follow announcements rather than ordinary demand.
At a glance
- Typical duration
- 30 minutes to 3 hours of degradation
- Usual cause
- Capacity, rate limiting, or a model rollout
- Tell-tale sign
- Interface responsive, answers slow or truncated
- Hidden reach
- Breaks products with no visible AI feature
What to do when ai tools is down
- Distinguish slow from down before reporting. A response that takes ninety seconds is a degradation. Knowing which one you are seeing changes what you should do.
- Do not retry aggressively. Automatic retries against a rate-limited service are the main reason these incidents last as long as they do.
- Check whether the API and the chat interface differ. They often have separate capacity, and one commonly works when the other does not.
Still not sure whether it is them or you? Check the site from our servers - We are on a different network, a different resolver and a different route from you.