How DeepSeek API pricing works
DeepSeek bills per token like other providers, with two twists: it prices input differently depending on whether it was served from cache, and it charges half price outside peak hours. There are two current models:
- DeepSeek V4 Pro (model ID
deepseek-v4-pro): $1.32 / $3.96 at peak, half that off-peak. - DeepSeek V4.1 Flash (model ID
deepseek-flash): $0.30 / $1.20 at peak, half that off-peak.
Both have a 1M-token context window and can write up to 384K tokens in one response. Both support thinking and non-thinking modes, tool calls and JSON output, and both accept requests in either the OpenAI or the Anthropic API format. V4.1 Flash also accepts images; V4 Pro doesn't.
The older model names deepseek-v4-flash and deepseek-v4-flash-vision-exp still work, but those models have been retired: requests to them are served by V4.1 Flash and billed at the Flash price.
Peak and off-peak hours
Off-peak rates are exactly half the peak rates, for input, cached input and output alike. DeepSeek's peak hours are:
| Peak period (UTC) | New York (EDT) | London (BST) | Singapore |
|---|---|---|---|
| 01:00–04:00 | 21:00–24:00 (previous day) | 02:00–05:00 | 09:00–12:00 |
| 06:00–10:00 | 02:00–06:00 | 07:00–11:00 | 14:00–18:00 |
Peak pricing applies Monday to Friday only, excluding Chinese public holidays. All other hours are off-peak, including weekends and Chinese public holidays in full. Local times above apply until clocks change on 25 October in the UK and 1 November in the US, after which peak hours start an hour earlier locally.
In practice, peak hours cover the Asian working day and the early European morning. The whole US working day is off-peak, as are European afternoons. A batch job scheduled for a weekend, or for 10:00–01:00 UTC on a weekday, pays half price.
Cache hits
DeepSeek caches repeated prompt prefixes automatically. Input served from cache costs $0.044 per 1M tokens on V4 Pro and $0.006 on V4.1 Flash at peak, and half that off-peak: about 3% and 2% of the normal input price. For a coding agent with 750M input tokens a month, 85% of it cached, and 8M output, V4 Pro costs about $208 a month at peak prices, and V4.1 Flash about $47.18.
DeepSeek doesn't list a separate batch discount; the off-peak price does the same job for work that can wait.
Is there a free tier?
DeepSeek's API pricing page lists no free allowance; API usage is billed per token from the first request.
Which DeepSeek model should you use?
- Coding, agents and reasoning: V4 Pro, one of the cheapest frontier-tier models we track.
- High-volume work, and anything with images: V4.1 Flash, at $0.30 / $1.20 per 1M tokens at peak.
- Batch jobs: either model, scheduled off-peak for half price.
Other things to know
- Concurrency limits: DeepSeek caps concurrent requests at 2,500 on V4.1 Flash and 500 on V4 Pro.
- Search and Fetch: DeepSeek’s own web search and page-fetch APIs are billed per request, at $0.75 per 1,000 successful requests each, on top of the tokens the model uses.
- Third-party hosts also serve DeepSeek models at their own prices. This page covers DeepSeek's own API.
Source: DeepSeek models and pricing. Checked 6 November 2026.
