AI Tools Status
Everything is up.
AI tools fail differently from ordinary web apps, and the difference matters when you are trying to work out whether the problem is yours. The editor, the chat window, or the app shell is usually served from normal web infrastructure and stays up, while the model behind it runs on separate, capacity-constrained hardware. So the interface loads, you type, and nothing comes back. That is why an AI tool can feel completely broken while every status page involved still reports green.
Capacity is the usual culprit rather than a hard outage. When demand spikes, requests queue, time to first token climbs, and providers shed load by returning rate-limit or server-busy errors. A provider may also degrade one model family while the rest answer normally, so the same product can be broken for one task and fine for another.
We check the endpoint each tool actually depends on, which for several of them means an unauthenticated API path rather than the marketing site. A few block automated requests outright, and for those we report unverifiable and lean on the vendor's own status page rather than guessing.
- up
- slow
- down
- no reading
ChatGPTUp
ClaudeUp
CohereUp
CursorUp
DeepSeekUp
ElevenLabsUp
Fireworks AIUp
Google GeminiUp
GrokUp
GroqUp
Hugging FaceUp
MidjourneyUp
Mistral AIUp
ModalUp
OpenAIUp
PerplexityUp
ReplicateUp
RunwayUp
SunoUp
Together AIUp
WindsurfUp
Common questions about AI tools
The interface and the model run on different infrastructure. A status page usually tracks the platform, and a model that is out of capacity returns errors or queues requests without the platform ever going down. Vendors also tend to post an incident only once it is widespread, so a short capacity squeeze often never appears at all.
Check whether the failure is the same across two different models from the same provider. A key or quota problem fails identically everywhere and usually returns a 401 or 429 with a quota message, while a capacity problem is often confined to one model and clears within minutes. If a second model works, the provider is up and the issue is your request or your account.
It means the service blocks our automated checks, or its only endpoint needs a key, so a failed request tells us nothing about whether the product works. We refuse to read that as an outage. Where the vendor publishes a machine-readable status page we fall back to it and label the verdict as theirs.