DeepSeek V4-Pro Explained: Peak and Off-Peak Pricing
In short: DeepSeek V4-Pro is the larger model in DeepSeek's V4 family, generally available since 13 August 2026. It costs $1.32 per million input tokens and $3.96 per million output at peak times, and half that off-peak. It has a 1M-token context window, can write up to 384K tokens in one reply, reads text only, and its weights are published under the MIT licence.
What is DeepSeek V4-Pro?
DeepSeek V4-Pro is a large language model from DeepSeek, a Chinese AI company, available through DeepSeek's API, its chat app and as downloadable open weights. It reached general availability on 13 August 2026; the API model name is deepseek-v4-pro, and DeepSeek's pricing page lists the version as DeepSeek-V4-Pro-0813. Its smaller sibling is covered in DeepSeek V4-Flash explained.
For how models like this work, see how LLMs work.
How much does DeepSeek V4-Pro cost?
DeepSeek charges by time of day: off-peak rates are half the peak rates. From DeepSeek's pricing page:
Peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, excluding Chinese public holidays. All other hours are off-peak, including weekends and Chinese public holidays in full. There is no separate endpoint or flag - the rate depends on when the request arrives. The whole 1M-token window is billed at the same rate, with no long-context surcharge.
What is peak and off-peak pricing?
It is pricing that changes with the time of day, like off-peak electricity. DeepSeek introduced it on 16 August 2026. A request sent during the listed peak windows pays the full rate; the same request sent at any other time pays half. Work that does not need an instant answer - large batches, evaluations, bulk summaries - can be scheduled for off-peak hours.
What are DeepSeek V4-Pro's specs?
| Spec | DeepSeek V4-Pro |
|---|---|
| Developer | DeepSeek |
| Generally available | 13 August 2026 |
| Model name | deepseek-v4-pro |
| Context window | 1M tokens |
| Maximum output | 384K tokens |
| Input / output | Text in, text out - vision is listed as not supported |
| Thinking | Thinking (default) and non-thinking modes; effort levels low, high and max |
| Features | JSON output, tool calls, Responses API and Anthropic API formats |
| Weights | Open weights on Hugging Face, MIT licence |
Can you download DeepSeek V4-Pro?
Yes - its weights are published on Hugging Face under the MIT licence, which permits commercial use. Running a model this size needs data-centre hardware; see how to run AI models locally for the software involved.
How does DeepSeek V4-Pro rank?
On LMArena's text leaderboard DeepSeek V4-Pro has 1458 Elo (rank 65). DeepSeek V4.1-Flash has 1476 Elo (rank 35, max effort). The full ranked table is on our benchmarks page.
Words to know
- General availability (GA) - a model's full release, after any preview or beta.
- Peak / off-peak pricing - rates that depend on the time a request is sent.
- Cache hit - input DeepSeek has already processed, billed at a lower rate.
- Thinking mode - the model reasons before answering, using extra output tokens.
- Open weights - model files published for anyone to download and run.
- MIT licence - a permissive licence that allows commercial use.
Related guides
Frequently asked questions
How much does DeepSeek V4-Pro cost?+
At peak times DeepSeek V4-Pro costs $1.32 per million input tokens and $3.96 per million output tokens. Off-peak, the rates halve to $0.66 and $1.98.
When is DeepSeek off-peak?+
DeepSeek's peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, excluding Chinese public holidays. Every other hour, including all of Saturday and Sunday, is billed at the off-peak rate.
What is DeepSeek V4-Pro's context window?+
DeepSeek V4-Pro has a 1M-token context window, billed at a flat rate, and can write up to 384K tokens in one reply.
Does DeepSeek V4-Pro support images?+
No. DeepSeek's pricing page lists vision as not supported for V4-Pro; it takes text input and returns text.
Is DeepSeek V4-Pro open source?+
Its weights are published on Hugging Face under the MIT licence, so it can be downloaded, run and used commercially.
Enjoyed this? Get more in your inbox.
Up to ten of the week's AI news stories, by email once a week.