Kimi is the AI assistant from Moonshot AI, a Chinese lab best known for handling very long documents and conversations. People use it for chat, deep research, web search, file analysis, slide and website generation, and programming through Kimi Code, via the web app at kimi.com, mobile apps, and a developer API on the Moonshot Open Platform. An outage often stalls long research or coding sessions mid-task. Start with the live verdict at the top of this page to see whether the problem is widespread right now.
Common reasons Kimi stops working
- Overload errors that are really rate limits. Kimi often returns a 429 “engine is currently overloaded” message that is actually a per-plan rate limit. It is notorious on Kimi for Coding, where clients retry automatically and stay stuck in a loop.
- Agent and coding backend incidents. The agentic component that powers agent tasks and coding runs has a history of frequent critical incidents. When it fails, long tasks die mid-run while plain chat keeps answering.
- SMS login codes that never arrive. Phone-based sign-in depends on verification codes, and delivery failures have locked users out entirely, including a confirmed sign-in incident in June 2026, even while the models stayed healthy.
- In-chat web search failures. The built-in search tool can error or return nothing while the model itself still responds. One incident in July 2026 pushed search error rates above 20 percent.
- Missing chat history or projects. Past conversations sometimes fail to load, often on the mobile app only while the web app is fine, or the reverse.
- Free-tier congestion at peak hours. “Server busy” messages and slow replies cluster around peak demand, and entry-level API accounts allow only about three requests per minute with one concurrent request.
How to fix Kimi when it is not working
- Check the status at the top of this page first. If Moonshot’s status feed, our own check, and user reports all show trouble, the problem is on Kimi’s side and waiting is the fastest fix.
- Stop retrying “engine overloaded” errors. That message usually means a plan rate limit. Pause for several minutes, and disable automatic retries in coding clients so you do not stay stuck in a loop.
- Reload and start a fresh chat. A refresh or a new conversation clears many stuck or half-finished responses.
- Use a different sign-in method. If an SMS code never arrives, sign in on kimi.com with email or Google instead of waiting on phone verification.
- Switch between the web app and the mobile app. History and loading problems often affect only one surface, so the other usually still works.
- Ask without search or agent features. If web search or an agent run keeps erroring, rephrase the request as a plain chat message. The core model is often still up.
- Try a private window or another network. Clearing cache, disabling extensions, or moving to mobile data rules out a local cause.
For a fuller checklist that applies to any AI service, see our guide on what to do when an AI tool is down.
Is Kimi down, or is it just you?
Compare your experience with what others are reporting. If the verdict at the top of this page shows a spike, the outage is real and shared. If reports are quiet but Kimi still fails for you, the cause is more likely local: your login method, a plan rate limit, your browser, or your network. Trying Kimi on a second device or network is the fastest single test, and an “overloaded” error on a quiet day usually points to your plan’s limits rather than an outage.
How we check Kimi
We determine Kimi’s status from two signals rather than one. Moonshot AI publishes an official status page with a machine-readable feed that breaks Kimi into separate components, and we read that feed directly. We also run our own independent reachability check against Kimi from the outside on a regular interval. Combining both means we catch problems even when one source is slow to update, and it keeps the verdict honest. You can read more about how this works on our methodology page.
Alternatives to Kimi during an outage
If Kimi is down and you need to keep working, a few comparable assistants cover most of the same ground. DeepSeek is a strong stand-in for reasoning and coding tasks, Qwen handles long documents and general assistant work well, and ChatGPT covers writing, research, and everyday questions. Confirm your chosen backup is healthy before switching, since AI tools sometimes fail together during broad infrastructure events.
When Kimi breaks one piece at a time
Moonshot runs Kimi as many separate components, with distinct model backends for text, vision, thinking, research, and agentic work, plus web search, the coding product, and the Open Platform API. They fail independently, so the typical Kimi outage is partial: agent and coding runs die while ordinary chat answers normally, or search fails inside a conversation that is otherwise working. The most misleading symptom of all is the “engine overloaded” error that is really a plan rate limit, not downtime.
Kimi also runs as two parallel deployments. The international app at kimi.com uses email or Google sign-in, while the Chinese domestic version at kimi.moonshot.cn uses phone-number login, and the .cn address redirects international visitors to kimi.com. SMS code failures, login problems, and support delays, which grow during Chinese public holidays, do not always match what the status page shows.
- Agent or coding runs failing while plain chat answers normally
- “Engine overloaded” errors that are really plan rate limits
- In-chat web search erroring while the model still responds
- Chat history missing on mobile while the web app works