DeepSeek V4 Pro

What DeepSeek says. The pricing page lists V4 Pro next to Flash as a feature and price table and does not describe target use cases for either. The Thinking Mode guide explains that thinking mode produces chain-of-thought reasoning before the final answer to improve accuracy, and that the model can run multiple rounds of reasoning and tool calls before answering.

Our editorial judgment. Because the two models share context, output limit and features, the reason to pay for Pro is quality on your hardest tasks, and DeepSeek publishes nothing on this page that tells you how large that gap is. We would treat Pro as an escalation tier: route difficult refactors, multi-step research or long tool-calling sessions to it, and keep routine traffic on Flash. Pro cannot take images, so any vision step has to go elsewhere. If you want a packaged coding assistant rather than a raw API, see our AI coding tools guide.

• No vision. Pipelines that read screenshots or scanned pages need another model for that step.
• Lower concurrency. The limit is 500 concurrent requests per account, regardless of API key. Above it you get HTTP 429. DeepSeek offers capacity expansion at no extra cost on request.
• Peak hours double the price. Peak is 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, excluding Chinese public holidays. All other hours, including weekends, are off-peak at half price.
• Version behind the name can change. The model name maps to DeepSeek-V4-Pro-0813 today. DeepSeek has already repointed retired Flash names to a newer model, so log the version and rerun your tests when it changes.
• Thinking is on by default. Reasoning tokens are produced unless you disable thinking, and they add to output cost.
• Ignored sampling settings. In thinking mode, temperature, presence_penalty and frequency_penalty have no effect, and top_p is clamped to 0.95 to 1.0.
• Tool calls need reasoning passed back. With tools in the request, omitting earlier reasoning_content returns a 400 error.
• Long waits. If inference has not started after 10 minutes, the server closes the connection.
• Prices can move. DeepSeek states it reserves the right to adjust prices.

Both share the 1M context, 384K maximum output, thinking modes, JSON output, tool calls and both API formats. At list rates Pro costs 4.4 times as much on cache-miss input, 3.3 times as much on output. Outside DeepSeek, Google lists Gemini 3.8 Flash at USD 0.75 input and USD 3.75 output per 1M tokens through 2026, rising to USD 1.50 and USD 7.50 in 2027, and Gemini 3.5 Flash-Lite at USD 0.30 and USD 2.50.

1. Collect 20 to 50 hard tasks from your own work, especially ones where Flash or another model already failed.
2. Run each on Pro and on V4.1 Flash with the same prompt, tools and thinking effort, passing reasoning_content back correctly in tool loops.
3. Score pass or fail against your rubric and log cache hits, misses and output tokens.
4. Compute cost per completed task at both peak and off-peak rates rather than cost per call.
5. Send a task type to Pro only if its pass rate gain covers the price gap, and check that peak traffic stays under 500 concurrent requests.

We read DeepSeek's Models & Pricing page, Thinking Mode guide and Rate Limit & Isolation page on 15 October 2026. We have not run DeepSeek V4 Pro, and nothing here is a test result or ranking. Prices and limits change; check DeepSeek's pages before committing budget. See our ranking method for how HumanRank scores tools and the model directory for other profiles.

What is DeepSeek V4 Pro? It is the higher-priced of DeepSeek's two API models, called as deepseek-v4-pro and served by the version DeepSeek-V4-Pro-0813.

How much does DeepSeek V4 Pro cost? It costs USD 1.32 per 1M cache-miss input tokens and USD 3.96 per 1M output tokens at peak, and half that off-peak.

Does DeepSeek V4 Pro support images? No, Vision is listed as not supported for V4 Pro; DeepSeek V4.1 Flash supports it.

What is the DeepSeek V4 Pro context length? It is 1M tokens, with a maximum output of 384K tokens.

What is the DeepSeek V4 Pro rate limit? It is 500 concurrent requests per account, and requests above that receive HTTP 429.

Is DeepSeek V4 Pro worth it over V4.1 Flash? Only if your tests show it completes tasks Flash cannot, since DeepSeek lists Flash at a lower price on every line with the same context and output limits.