Pricing
DeepSeek API pricing: V4.1 Flash, V4 Pro, off-peak hours and cache
What each DeepSeek model costs per token, when the half-price hours are, how cache hits work, the limits, and how it compares with Claude and OpenAI.
DeepSeek's API has two models. DeepSeek V4.1 Flash (deepseek-flash) costs $0.30 per million input tokens and $1.20 per million output, and DeepSeek V4 Pro (deepseek-v4-pro) $1.32 and $3.96. Input the cache has already seen costs $0.006 on Flash and $0.044 on Pro. Outside peak hours, which are 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays, everything is half price, which covers the whole US working day. Both models have a 1M-token context, and there's no subscription: you top up a balance and pay per token. Its web search and page-fetch tools cost $0.75 per 1,000 calls.
Key takeaways
- V4.1 Flash: $0.30 input, $1.20 output, $0.006 cached; off-peak $0.15, $0.60, $0.003.
- V4 Pro: $1.32 input, $3.96 output, $0.044 cached; off-peak $0.66, $1.98, $0.022.
- Off-peak: everything outside 01:00–04:00 and 06:00–10:00 UTC on weekdays, plus weekends and Chinese public holidays.
- Limits: 2,500 requests at once on Flash and 500 on Pro per account; raising them is free on request.
- Claude Code: works through DeepSeek's Anthropic-compatible endpoint, billed at these prices.
| Model | Input, cache hit | Input, cache miss | Output | Off-peak |
|---|---|---|---|---|
| DeepSeek V4.1 Flash (deepseek-flash) | $0.006 | $0.30 | $1.20 | Half: $0.003 / $0.15 / $0.60 |
| DeepSeek V4 Pro (deepseek-v4-pro) | $0.044 | $1.32 | $3.96 | Half: $0.022 / $0.66 / $1.98 |
| Search or Fetch API | — | — | $0.75 per 1,000 calls | Same |
How much does the DeepSeek API cost?
From DeepSeek's pricing page (DeepSeek), per million tokens:
- DeepSeek V4.1 Flash (
deepseek-flash, the October 17 build): input $0.30 on a cache miss, $0.006 on a cache hit; output $1.20. - DeepSeek V4 Pro (
deepseek-v4-pro, the August 13 build): input $1.32 on a miss, $0.044 on a hit; output $3.96.
Both think by default, and thinking tokens bill as output. The context window is 1M tokens and a reply can run to 384,000 tokens. DeepSeek cut prices when V4.1 Flash arrived on September 10, 2026, and left them unchanged with the October 17 update. The old deepseek-chat and deepseek-reasoner names stopped working on July 24, 2026.
When are DeepSeek's off-peak hours?
Since August 16, 2026, DeepSeek charges half price outside peak hours: "Off-peak rates are half of the peak rates." Peak is 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday. Everything else is off-peak, including weekends and Chinese public holidays.
In US time, peak is 9 p.m. to midnight and 2 a.m. to 6 a.m. Eastern on weeknights (an hour earlier in winter), so a US team working daytime hours pays the off-peak rate nearly all the time. Europe's morning overlaps the second peak window.
How does DeepSeek's cache pricing work?
When a request starts with text DeepSeek has recently seen, that part bills at the cache-hit rate: $0.006 per million on Flash, 2% of the normal input price. Coding agents resend the whole conversation on every step, so most of their input is cache hits; on DeepSeek that makes long sessions very cheap. There's no separate charge for writing to the cache.
What else does the DeepSeek API charge for?
- Search and Fetch APIs: $0.75 per 1,000 successful requests each.
- Free credit: DeepSeek's terms mention a free quota without an amount; granted balance is used before money you've topped up.
- Rate limits: 2,500 concurrent requests on Flash and 500 on Pro per account. Above that you get HTTP 429, and DeepSeek raises limits free on request (DeepSeek).
How does DeepSeek compare with Claude and OpenAI on price?
Per million input and output tokens:
- DeepSeek V4.1 Flash: $0.30 and $1.20, half off-peak.
- Claude Haiku 5.5 and GPT-6 Luna: $0.10 and $0.50 each, cheaper per token than DeepSeek Flash at peak.
- DeepSeek V4 Pro: $1.32 and $3.96.
- Claude Sonnet 5.5 and GPT-6.1 Sol: $2 and $10.
- Claude Opus 5.5: $4 and $20. GPT-6 Astra: $10 and $50.
The token price isn't the whole cost. On Vals AI's Terminal-Bench 4.0 run, a task cost $0.54 on V4.1 Flash, $3.31 on V4 Pro and $13.20 on Claude Opus 5.5, but the DeepSeek models solved far fewer tasks: 31.8% and 14.1% against 65.2% (DeepSeek V4 vs Claude). Full price lists: Claude API pricing, OpenAI API pricing.
Can you use the DeepSeek API with Claude Code?
Yes. Point Claude Code at https://api.deepseek.com/anthropic with ANTHROPIC_BASE_URL, put your DeepSeek key in ANTHROPIC_AUTH_TOKEN, and DeepSeek recommends deepseek-flash[1m] as the model (DeepSeek). Any model name beginning claude-opus is billed as V4 Pro, and other Claude names as Flash. DeepSeek documents Codex, OpenCode and other agents the same way. Anthropic doesn't support running Claude Code on another provider.
Where does your data go?
DeepSeek's privacy policy says it stores personal data in China and uses input to improve its models, with an opt-out by email; its API terms don't say whether API data is used (DeepSeek). The US has ordered DeepSeek products out of federal civilian agencies, published October 20, 2026. The models' weights are MIT-licensed, so running them on your own servers or a cloud provider avoids DeepSeek's API, at that provider's prices. More in DeepSeek V4 vs Claude.
Models without an API account
Poly (usepoly.co) doesn't offer DeepSeek. It runs Claude, GPT, Grok and Muse for a team in one shared room, billed as credits per person with no API keys, and every account starts with free models from each lab. What is Poly?
Common questions
How much does the DeepSeek API cost?
DeepSeek V4.1 Flash costs $0.30 per million input tokens and $1.20 per million output; DeepSeek V4 Pro costs $1.32 and $3.96. Cached input is $0.006 and $0.044, and everything is half price off-peak.
When is DeepSeek off-peak?
Everything outside 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays, plus weekends and Chinese public holidays. Off-peak prices are half the peak prices, which covers the US working day.
Is the DeepSeek API free?
No. It is pay as you go from a topped-up balance. DeepSeek's terms mention a free quota without an amount, and its open-weight models are free to download and run yourself under the MIT license.
What happened to deepseek-chat and deepseek-reasoner?
They were retired on July 24, 2026. The current API names are deepseek-flash, for V4.1 Flash, and deepseek-v4-pro, for V4 Pro, and both can think or answer directly.
Is DeepSeek cheaper than Claude?
Than Sonnet and Opus, yes, by roughly 7 to 17 times per token. Claude Haiku 5.5, at $0.10 and $0.50 per million, is cheaper per token than DeepSeek V4.1 Flash at peak.