AI status
AI status: is it down, or is it just you?
Is ChatGPT down, or Claude, or Gemini? Every 5 minutes we send a tiny test message to each major AI family, the way llmwise's chat sends them, and read each maker's own status page. Here's what both say now.
Updated . While people are reading it, this page is refreshed about once a minute.
Right now
Our verdict is each family's last three tests: the last 15 minutes, and Grok's last 45. The official status is each maker's own page.
| Service | Our tests | Answered, 24 h | First token | Official status |
|---|---|---|---|---|
| ChatGPT | Answering | 100% | 0.8 s | Operational (status.openai.com) |
| Claude | Answering | 100% | 1.0 s | Operational (status.claude.com) |
| Gemini | Answering | 100% | 1.6 s | No open incidents (Google Cloud and Workspace status) |
| Grok | Answering | 100% | 0.9 s | No open incidents (status.x.ai) |
| DeepSeek | Answering | 100% | 0.4 s | No public status feed |
| Kimi | Answering | 96.6% | 0.9 s | Operational (status.moonshot.cn) |
| GLM | Answering | 100% | 0.7 s | No public status feed |
| The OpenAI API | Answering | 100% | 0.8 s | Operational (status.openai.com) |
| The Anthropic API | Answering | 100% | 1.0 s | Operational (status.claude.com) |
How the tests work
Every 5 minutes, "Reply with OK" goes to each family's cheapest model (Grok's every 15: it thinks before every answer, so each test costs more), and every half hour to its most capable one too: Anthropic: Claude Haiku 4.5 and Claude Fable 5.1, OpenAI: GPT-6 Luna and GPT-6 Astra, Google: Gemini 3.8 Flash and Gemini 3.1 Pro (preview), xAI: Grok 4.7, DeepSeek: DeepSeek V4.1 Flash and DeepSeek V4 Pro, Moonshot: Kimi K3, and Z.ai: GLM 5.3 Flash and GLM 5.3.
Each goes the way llmwise's chat sends it: straight to the model's maker, or through OpenRouter to hosts that don't store or train on prompts.
The reply is capped at 32 tokens, and each test is tried once: no retry, no switch to another route.
A test fails when the model is too busy (a rate limit or "overloaded"), breaks (a server error), drops the connection, or takes more than 30 seconds. A test our own request broke doesn't count against anyone.
We keep the results for 30 days.
The apps are not the models
ChatGPT, claude.ai, the Gemini app and Grok are apps built on these models, with their own websites, sign-in and limits. When our tests fail, the models are struggling, and the apps usually are too. When only an app fails, its maker's status page is the one to watch: each page here shows it.
Questions
Is it down, or is it just me?
It's the question behind every page here. If our test of a model fails, the model's maker is having trouble for everyone who calls it, not just you. If our test passes and the maker's status page is clear, the problem is more likely your browser, your network, your account or its limits: each page lists the fixes for its product.
Why do you test the models through their APIs?
Because that's what apps like ChatGPT and Claude run on, and what llmwise's own chat uses: the models themselves. We don't automate other companies' websites. The apps can fail on their own (a sign-in page, a website), which is why every page also shows the maker's own status for its app.
How fresh is this?
Each page is refreshed at most a minute after its last visit, and new tests run every 5 minutes, so what you see is at most a few minutes old. Every page says when it was last updated.
Do the tests use anyone's messages?
No. The tests send only "Reply with OK". We keep each test's result (whether it answered, how it failed, how long it took, and which route it took) for 30 days, and nothing else.
Claude, GPT, Gemini, DeepSeek, Grok, Kimi, and GLM, in one chat.
See what a message costs before you send it. Free is 5 messages to try; sign in with an email link, no password or card.