DeepSeek Not Working
Seeing “service unavailable”, slow responses, or API errors? DeepSeek frequently hits capacity limits during peak demand. Here’s how to diagnose and work around it.
DeepSeek — live status
DeepSeek often shows Degraded rather than Down during high traffic — capacity issues cause errors even when the service is online. View incidents →
Known pattern: DeepSeek experiences recurring capacity spikes during Chinese national holidays, after major model releases, and during US evening hours (8–11 PM ET). During these windows, “service unavailable” errors are common even though the underlying model is running.
What’s happening?
“Service unavailable” or “Server busy”
This is DeepSeek’s capacity error — not a full outage. The queue is full and the server is rejecting new requests. Wait 5–15 minutes and retry, or switch to a different DeepSeek model (V3 vs R1 use separate pools). OpenRouter provides a load-balanced alternative.
API authentication errors (401 / Invalid API key)
The DeepSeek API (api.deepseek.com) is completely separate from the chat interface (chat.deepseek.com). Your API key is managed at platform.deepseek.com — a chat account does not automatically give you API access. Generate an API key at platform.deepseek.com/api_keys.
Very slow responses
DeepSeek R1 is a large reasoning model — its chain-of-thought process is inherently slower than chat models. If speed matters, switch to DeepSeek V3 or DeepSeek R1-zero (lighter reasoning variant). During peak load, even fast models can slow significantly.
Site not loading — possible geographic block
Some ISPs and enterprise networks block deepseek.com. If the site does not load at all (no error message, just timeout), your network may be filtering it. Try accessing via a different network, or use DeepSeek models through OpenRouter or Azure AI which serve the models from different domains.
4 fixes to try in order
Wait and retry — capacity fluctuates fast
DeepSeek’s capacity issues are usually temporary. During peak demand, availability can swing from “service unavailable” to fully working within 5–10 minutes as load shifts. If you see a capacity error, wait a few minutes before assuming the service is down.
Switch between DeepSeek model variants
On chat.deepseek.com, click the model selector and switch between DeepSeek V3 and DeepSeek R1. These models run on separate server pools, so one may have available capacity when the other does not. For API users, the model IDs are deepseek-chat (V3) and deepseek-reasoner (R1).
Access DeepSeek via OpenRouter
openrouter.ai provides DeepSeek V3 and R1 with load balancing across multiple hosting providers. When DeepSeek’s own infrastructure is saturated, OpenRouter’s routing often remains available. It also routes around geographic restrictions on deepseek.com.
Use DeepSeek on Azure or AWS instead
Microsoft Azure AI Foundry and AWS Bedrock host DeepSeek models on their own infrastructure. These deployments have dedicated capacity and are unaffected by traffic spikes on deepseek.com. They are also not subject to the same geographic restrictions and are a reliable fallback for production workloads.
chat.deepseek.com vs platform.deepseek.com — two separate systems
DeepSeek runs two distinct services that do not share accounts, quotas, or authentication:
- chat.deepseek.com — the free consumer chat interface. Uses your DeepSeek account login. Has its own capacity pool. Free to use with a DeepSeek account.
- platform.deepseek.com — the developer API platform. Requires a separate API key from platform.deepseek.com/api_keys. Uses token-based billing. Has its own separate capacity and rate limits.
If the chat interface is down but the API is up (or vice versa), that is expected — they are independent systems. Check both via the status badge above and the DeepSeek status page.
Get alerted when DeepSeek capacity recovers
Star DeepSeek on Prismix for a free email when service recovers from capacity overload. Also monitors ChatGPT, Claude, and 70+ other AI services.