DeepSeek not working: what each error means and how to fix it

If DeepSeek is not working, check DeepSeek's status page first, then match what you see to the error code. DeepSeek's docs give a fix for each API error: 503 means the servers are overloaded and you should wait and retry, 429 means you are sending too many requests at once, and 402 means your balance is empty. This guide covers what the official sources say, as of November 1, 2026.

One limit up front. DeepSeek's docs describe errors for the API. We found no official help page for the "server is busy" message in the chat app, so the app section below sticks to what the status page shows.

Check the DeepSeek status page first

status.deepseek.com lists each service separately: the V4 Pro API, the V4.1 Flash API, the Search and Fetch APIs, and the chat services, which include file upload and search. As of November 1, 2026 it reported "Everything is running smoothly" and listed no recent incidents.

The page also shows uptime over the last three months. For August to November 2026 it listed 99.74% for the V4 Pro API and 99.47% for the V4.1 Flash API. If the service you use shows an open incident, nothing on your side needs fixing. Wait for DeepSeek to resolve it. The page has a "Subscribe to updates" option, though it does not explain how that works.

What each DeepSeek API error means

The error codes page lists seven errors.

The two that readers search for most are 503 and 429, and they have different causes. A 503 is on DeepSeek's side and passes on its own. A 429 is about how many requests you have in flight, so your code is what changes.

Fix "server busy" and 503 errors

The docs say one thing about a 503: wait a short time and retry. Nothing in your key, balance or request needs to change.

If your client calls the API in a loop, add a pause between retries instead of sending them back to back. DeepSeek's docs do not publish a retry schedule, so pick your own delay and keep it short enough that a user is not left waiting.

Check the status page before you spend time debugging. A 500 or 503 across all your requests, with no change on your side, points to the service.

Fix 429 rate limit errors

The rate limit page describes limits on concurrent requests, not on requests per minute. As of November 1, 2026, an account can have up to 2,500 requests in flight for deepseek-flash and 500 for deepseek-v4-pro. A request counts from the moment it is sent until the model finishes its response.

Two details from that page matter in practice:

• The limit applies to the account. Creating more API keys does not raise it.
• Requests over the limit get a 429. Requests under it are answered normally.

So the fix is to cap how many requests your code runs at once. DeepSeek says higher limits are available through a capacity expansion request, at no additional cost.

Slow responses and dropped connections

While your request waits, DeepSeek keeps the connection open. Non-streaming requests receive empty lines, and streaming requests receive : keep-alive comments. They do not affect the JSON body, but a hand-written HTTP parser has to skip them. If inference has not started after 10 minutes, the server closes the connection.

Fix 401 and 402 errors

A 401 means the key is wrong. Check that you copied the whole key and that the Authorization header reads Bearer followed by the key. The API quick start uses https://api.deepseek.com as the base URL and the DEEPSEEK_API_KEY environment variable in its examples. If you do not have a key yet, see our guide on how to get a DeepSeek API key.

A 402 means the balance is empty. Top up on the DeepSeek platform and send the request again. If you are not sure what the API costs, is DeepSeek free covers it.

Check your model name

The quick start lists two current model names: deepseek-flash and deepseek-v4-pro. It says the older deepseek-v4-flash name is still accepted and goes to DeepSeek-V4.1-Flash-1017 at the Flash price.

DeepSeek's V4 announcement also said the older deepseek-chat and deepseek-reasoner names would be retired after July 24, 2026, 15:59 UTC. If an old integration stopped working, check which model name it sends. The docs do not say which error a retired name returns, so we can't tell you what to look for in the response.

If the chat app is the problem

The docs we read do not cover the chat app's "server is busy" message, so we can't promise a fix from DeepSeek. What you can do:

1. Open the status page and look at the chat service, file upload and search components.
2. If one shows an incident, wait for it to clear.
3. If everything is green, the problem may be local to you. The docs have no steps for this.

For account or API problems that continue, DeepSeek's docs list api-service@deepseek.com as a contact. For how DeepSeek's models compare with others, see is DeepSeek open source, which also explains which weights you can run yourself.