DeepSeek Releases V4-Pro for Complex Agentic Tasks
DeepSeek has introduced DeepSeek-V4-Pro — a new model with significantly improved performance in agentic scenarios. For V4-Pro and V4-Flash, users can choose the level of computational effort: low for simple queries, high for standard workflows, and maximum for the most complex tasks.
V4-Pro is already available through the app, website, and API. The company has also added native support for the OpenAI Responses API and quick setup for Codex. The API model names have not changed, and the connection parameters are available in the documentation.
At the same time, DeepSeek is revising its API pricing and introducing separate peak and off-peak rates. Rates will be 50% lower during off-peak hours; the new prices will take effect on August 16, 2026, at 16:00 UTC.
Why it matters
- —Adjustable reasoning levels make it possible to balance task complexity with speed and cost.
- —Compatibility with the OpenAI Responses API simplifies integrating the model with existing tools and agents.
- —A 50% discount during off-peak hours helps developers reduce costs for tasks that do not require an immediate response.
Key facts
- V4-Pro delivers improved performance in agentic scenarios.
- Low, high, and maximum compute effort levels are available for V4-Pro and V4-Flash.
- The model is available in the app, on the website via Expert Mode, and through the API.
- API model names remain unchanged.
- The new pricing will take effect on August 16, 2026, at 16:00 UTC.
The full text is in the original source. Here we provide a brief summary and key facts.