DeepSeek API Free Credits & Pricing (2026): Is DeepSeek API Free?

Short answer: the DeepSeek API is not free, but it is one of the cheapest frontier-class APIs we track. DeepSeek's official pricing page publishes no signup credit or free tier. What it does publish is low pay-as-you-go pricing, starting at $0.15 per million input tokens and $0.60 per million output tokens for deepseek-flash during off-peak hours, plus very cheap cached input. This page explains the prices, the off-peak discount, the 'granted balance' you may see in your account, and how to keep costs close to zero.

Is there any free DeepSeek API credit?

DeepSeek's docs describe two kinds of balance. Fees are deducted from your topped-up balance (money you paid) or a granted balance, and when both exist the granted balance is used first. The docs never say who receives a granted balance, how much, or when it expires. Third-party sites claim new accounts get figures such as 5 million free tokens, but no DeepSeek page confirms that, so we list free credits as unconfirmed.

The only free usage DeepSeek documents right now is temporary: Search and Fetch tool calls made through DeepSeek Harness, Claude Code or Codex are free until 00:00 Beijing time on 1 November 2026 (16:00 UTC on 31 October). After that they cost $0.75 per 1,000 successful requests.

If you open the billing page after signing up and see a non-zero balance you did not pay for, that is a granted balance. Treat it as a bonus to test with, not something to plan around.

DeepSeek API prices (October 2026)

Cost is simply tokens × price. Output includes reasoning tokens when thinking mode is on, which is the default for both models, so a model that 'thinks' longer costs more per answer.

The two models

DeepSeek released DeepSeek-V4.1-Flash-1017 on 17 October 2026 and open-sourced its weights on Hugging Face and ModelScope. The release note says it scores better on benchmarks than the previous Flash version while finishing faster with shorter outputs. Older names `deepseek-v4-flash` and `deepseek-v4-flash-vision-exp` still work but are routed to the new model and billed at the Flash price.

Peak and off-peak hours

Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday, excluding Chinese public holidays. Everything else is off-peak at half the peak price, including all of Saturday and Sunday and full Chinese public holidays. In practice that means:

• On US Pacific daylight time (UTC-7), the peak windows fall at 18:00–21:00 and 23:00–03:00 local time, so normal Pacific working hours are off-peak but evening jobs may not be.
• European mornings overlap the 06:00–10:00 UTC window, so batch jobs started at 09:00 in Berlin may be billed at peak rates.
• Weekend batch jobs are always off-peak.

Worked cost examples

Assume a workload of 10 million input tokens (all cache misses) and 2 million output tokens:

Caching changes the picture. If 8M of those 10M input tokens are cache hits on deepseek-flash off-peak, input costs 8 × $0.003 + 2 × $0.15 = $0.324, and the total with output is $1.524, roughly 44% less than the uncached $2.70. Agents and chat apps that resend the same long system prompt or document benefit most.

For context, compare these rates with our pages on OpenAI API pricing, Claude API pricing and Gemini API pricing.

Per-conversation example

Take a support chatbot handling 1,000 conversations, each using about 3,000 input tokens and 500 output tokens, with no cache hits. That is 3M input and 0.5M output tokens in total. On deepseek-flash off-peak it costs 3 × $0.15 + 0.5 × $0.60 = $0.45 + $0.30 = $0.75, or under a tenth of a cent per conversation. On deepseek-v4-pro off-peak it costs 3 × $0.66 + 0.5 × $1.98 = $1.98 + $0.99 = $2.97. At peak, both figures double.

Search and Fetch API

Alongside the V4.1 Flash release, DeepSeek made its Search and Fetch APIs generally available. They run on DeepSeek's in-house search engine and let a model look things up on the web or read a given page. Pricing is per successful request rather than per token:

• $0.75 per 1,000 successful requests for each endpoint. For example, 10,000 searches cost 10 × $0.75 = $7.50, plus the tokens the model reads and writes.
• Calls made through server-side tools in DeepSeek Harness, Claude Code or Codex are free until per-request billing starts at 00:00 Beijing time on 1 November 2026.
• Failed requests are not charged; DeepSeek bills successful requests only.

Because the API accepts the Anthropic message format, developers can point coding tools that speak that format at `https://api.deepseek.com/anthropic`. If you want to compare that route with Anthropic's own tools, see our Claude Code pricing page.

Who DeepSeek is a good fit for

• High-volume text jobs such as classification, summarisation and extraction, where off-peak Flash pricing keeps costs in cents.
• Long-document work, since both models accept up to 1M tokens of context and cached input is very cheap.
• Teams already using OpenAI or Anthropic SDKs, who can switch with a base URL and model name change.
• Not a fit if you need a guaranteed free allowance for a class or hackathon; DeepSeek publishes none. Use a documented free tier instead.

How to start using the DeepSeek API

1. Create an account at platform.deepseek.com.
2. Open the billing page to check for any granted balance, then use Top up to add funds. DeepSeek's docs do not publish a minimum top-up amount or the accepted payment methods; the top-up page shows what is available to you.
3. Create a key on the API keys page and store it as `DEEPSEEK_API_KEY`.
4. Point any OpenAI-compatible client at `https://api.deepseek.com`, or an Anthropic-compatible client at `https://api.deepseek.com/anthropic`.
5. Use model `deepseek-flash` for most work and `deepseek-v4-pro` when you need the larger model.

curl https://api.deepseek.com/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY" \
  -d '{"model":"deepseek-flash","messages":[{"role":"user","content":"Hello"}],"stream":false}'

Limits and errors you will meet

• Concurrency, not RPM. DeepSeek counts concurrent requests per account across all keys: 2,500 for deepseek-flash and 500 for deepseek-v4-pro. A request occupies a slot from when it is sent until the response finishes.
• 429 Rate Limit Reached when you exceed concurrency. DeepSeek says you can submit a capacity expansion request, and there is no extra cost for it.
• 402 Insufficient Balance when your balance runs out. Top up to continue; there is no free fallback.
• 503 Server Overloaded under high traffic. DeepSeek's advice is to retry after a brief wait.
• Keep-alive output. While waiting, non-streaming responses send empty lines and streaming responses send SSE keep-alive comments. If inference has not started after 10 minutes, the server closes the connection.

How to keep DeepSeek costs near zero

• Schedule batch work off-peak. Moving a nightly job out of the two UTC peak windows halves the bill with no code change.
• Reuse prompt prefixes. Cache hits on deepseek-flash cost $0.003 per million tokens, 50 times less than a cache miss. Keep system prompts and shared context identical and at the start of the message list.
• Turn thinking off when you don't need it. Both models default to thinking mode; simple extraction or formatting tasks rarely need reasoning tokens.
• Start with Flash. deepseek-v4-pro costs about 4.4 times as much per input token and 3.3 times as much per output token off-peak.
• Watch the Search/Fetch switch-over. If you rely on these tools in Claude Code or Codex, expect charges from 1 November 2026.

Free alternatives if you need $0

If you need genuinely free API access rather than very cheap access, these documented free tiers are better starting points:

• Gemini API free tier: free Flash and 2.5 Pro models with no card, in exchange for data use.
• Groq API free tier: published free limits for open-weight models such as gpt-oss-120b.
• OpenRouter free models: free endpoints from several providers behind one API.
• NVIDIA NIM API for hosted open models. (Together AI no longer gives signup credits.)

See our full list of free AI API keys for more.

DeepSeek's pricing page says the company reserves the right to adjust prices and recommends checking the page regularly. The figures above were checked on 17 October 2026.