How to raise your DeepSeek API rate limits

What is the limit?

DeepSeek doesn’t publish requests per minute or tokens per minute. It limits concurrency, meaning how many requests your account has in flight at once.

Model	Concurrency limit
deepseek-flash	2,500
deepseek-v4-pro	500

A request counts from the moment you send it until the response finishes. Long streamed answers hold a slot the whole time. Go over the limit and you get a 429 - Rate Limit Reached.

The limit is per account. Every API key on it shares the same pool.

How do you ask for more?

1. Fill out DeepSeek’s capacity expansion request. It’s linked from the rate limit page.
2. Describe your real business needs. DeepSeek says it matches concurrency to them.
3. Wait to hear back. There’s no charge for the increase.

What changes after an increase?

DeepSeek starts limiting each user_id you pass, on top of the account total. Each user ID gets up to 2,500 concurrent requests on deepseek-flash and 500 on deepseek-v4-pro. An empty ID counts as one special user, so set user_id on every request or your unlabeled traffic all shares one slot pool.

You put it in the request body with the OpenAI format, under extra_body with the OpenAI SDK, or in metadata with the Anthropic format. Keep it to letters, digits, underscores and hyphens, under 512 bytes, and don’t put personal data in it.

Is it a rate limit or a balance problem?

A 429 is concurrency or request speed. A 402 means your balance ran out, and no capacity request will fix that. Here’s how to fix a DeepSeek 402.

DeepSeek runs on a prepaid balance. If you pay for top-ups by card, ernd gives 4% back on that AI vendor spend and 1.5% on everything else. It’s a credit card for US companies. Request the card at app.ernd.ai.

DeepSeek’s open-weight models are also sold by other providers, with their own limits. See DeepSeek V4.1 Flash prices by provider, or compare rate limits across vendors on our AI API rate limits page.