DeepSeek unveils V4.1-Flash-1017 with stronger performance and open APIs

DeepSeek published an announcement on 2026-10-17. The points below summarize the company's statements; performance claims have not been independently benchmarked by ChineseModels.

What was announced

• DeepSeek-V4.1-Flash-1017 delivers improved benchmarks, faster execution, and more concise responses than its predecessor, representing gains across efficiency and cost.

• The model is integrated with DeepSeek Harness, Tencent CodeBuddy, and WorkBuddy; upgrading to Harness v0.2.1-rc.1 or newer is advised for optimal use.

• Developed via reinforcement learning with self-feedback, the new model can build complete projects from scratch, including 3D scenes like a Forbidden City model and a racing game.

• Search and Fetch APIs, powered by an in-house search engine handling hundreds of millions of daily requests, are now public at $0.75 per 1,000 calls per endpoint.

• For DeepSeek API users, server-side tool calls for Search and Fetch are free until Oct 31, 2026, at 16:00 UTC, after which billing applies; the model is also open-sourced on Hugging Face and ModelScope.

What this means for integration

An announcement does not establish availability, pricing or input support in ChineseModels. Check the model catalogue and integration documentation, then test your application's exact request before switching models.

Source: DeepSeek-V4.1-Flash-1017: Wish-Driven Programming Made Real | DeepSeek API Docs. Retrieved 2026-10-22.

This is an automatically prepared, source-checked news summary. Report corrections to support@chinesemodels.ai.