DeepSeek V4 Pro: Official Features, Pricing, and How to Use It
DeepSeek V4 Pro is DeepSeek’s official flagship V4 model, released in general availability on 13 August 2026, with a 1 million token context window and separate peak and off-peak API prices.
What is DeepSeek V4 Pro?
DeepSeek V4 Pro is the official Pro model in DeepSeek’s V4 lineup. The company’s English announcement, DeepSeek-V4-Pro GA Release, says the general-availability model launched with stronger agent performance in production, more flexible reasoning effort, and native support for the OpenAI Responses API, including a one-click setup aimed at Codex. The same post says V4 Pro is available in the DeepSeek app and on the web through Expert Mode, and through the API under the existing model name.
The official Models & Pricing page lists the serving version as DeepSeek-V4-Pro-0813. The API model name is `deepseek-v4-pro`. The OpenAI-compatible base URL is `https://api.deepseek.com`; an Anthropic-compatible path is also documented at `https://api.deepseek.com/anthropic`.
Shared limits for the current V4 API models include a 1 million token context window and a maximum output of 384K tokens. Both Pro and Flash support JSON output, tool calls, the Responses API, the Anthropic API, and (in beta) chat prefix completion. Fill-in-the-middle completion is documented as beta and is limited to non-thinking mode. One important difference: vision is supported on Flash and is not supported on V4 Pro.
Community fine-tunes that reuse the “DeepSeek V4 Pro” name sometimes appear on Hugging Face. Those are third-party weights, not the official API model described in DeepSeek’s docs. If you want the product DeepSeek actually ships, use the app, the website, or the official API — not an unrelated repository title.
How to use it
In the app or on the web
DeepSeek’s announcement is explicit: choose Expert Mode in the official app or website to use the generally available V4 Pro model. If you stay on the default or non-expert path, you may be on a different V4 model (for example Flash).
Through the API
Send requests with the model name `deepseek-v4-pro` to the official base URL. The pricing page notes that thinking mode is supported and is the default; you can switch to non-thinking mode by following the thinking-mode guide in the same docs. For V4 Pro and V4 Flash, thinking intensity can be set to low, high, or max: the announcement recommends low for simple tasks, high for everyday agent work, and max for more complex jobs.
The GA post also highlights native Responses API support and a documented path for connecting Codex. Use DeepSeek’s own “quick start / agent tools / Codex” documentation rather than a third-party gist if you are wiring that up.
What not to expect
Do not send image inputs to V4 Pro and expect official vision support — the pricing table marks vision as unsupported on Pro. For image understanding, the same table points you to the Flash line instead. Also treat prices, peak hours, and concurrency caps as values that DeepSeek can change; the docs say to re-check the pricing page.
Official API pricing (as published)
DeepSeek bills per million tokens and, for the V4 official release, uses peak and off-peak rates. Off-peak rates are half of peak. The English announcement says the new prices took effect at 16:00 UTC on 16 August 2026.
Peak hours, from the official pricing page, are 01:00–04:00 and 06:00–10:00 UTC, Monday through Friday, excluding Chinese public holidays. All other hours, including weekends and Chinese public holidays in full, are off-peak.
For `deepseek-v4-pro`, the published rates per 1 million tokens are:
- Input, cache hit: $0.022 off-peak / $0.044 peak
- Input, cache miss: $0.66 off-peak / $1.32 peak
- Output: $1.98 off-peak / $3.96 peak
The same page lists a concurrency limit of 500 for V4 Pro, compared with 2500 for Flash. Search and Fetch, if you use those standalone APIs, are billed separately at $0.75 per 1,000 successful requests. Fees come out of your topped-up or granted balance, with granted balance used first when both exist.
Flash remains the cheaper, higher-concurrency option and is the one with vision. Pro is the official higher-priced flagship. Compare those published numbers rather than informal “times more expensive” claims, which can mix cache-hit and cache-miss rows.
Where to get it
Use Expert Mode in the official DeepSeek app or website, or call `deepseek-v4-pro` on the official API. Start from DeepSeek’s pricing docs and the GA announcement. The API platform page is the company’s own entry point for building on the V4 series.
FAQ
When was DeepSeek V4 Pro officially released?
DeepSeek published the general-availability announcement for V4 Pro on 13 August 2026. The API serving version listed in the docs is DeepSeek-V4-Pro-0813.
What is the API model name?
Use deepseek-v4-pro. The OpenAI-format base URL is https://api.deepseek.com.
Does DeepSeek V4 Pro support images?
No. Official docs say vision is not supported on V4 Pro. The Flash model in the same table does support vision.
How do I use V4 Pro in the app?
Open the official DeepSeek app or website and select Expert Mode, according to the GA announcement.
Are Hugging Face “DeepSeek V4 Pro” uploads the official model?
Not necessarily. Third-party fine-tunes can reuse the name. The official product is the app, website, and API documented by DeepSeek.
Sources: DeepSeek Models & Pricing · DeepSeek-V4-Pro GA Release · Chinese announcement (same release)
This is an unofficial guide. Pricing, limits, and features are taken from DeepSeek’s public docs as of the dates those pages were checked and can change. We are not affiliated with DeepSeek.