DeepSeek launched DeepSeek V4.1-Flash-1017 on Friday, October 17, 2026. The Flash model update is described by the company as a Pareto improvement in performance and cost: higher benchmarks, faster task execution and shorter outputs, all without changing the API price.
What changed in DeepSeek V4.1-Flash-1017
Version 1017 replaces the previous model on the website, app and API under the same deepseek-flash identifier. The weights of DeepSeek V4.1-Flash-1017 were released under the MIT license on Hugging Face and ModelScope. According to the model card, there are gains in reasoning, coding, agentic tasks and visual benchmarks.
Official numbers for DeepSeek V4.1-Flash-1017 include: Terminal-Bench 4.0 rose from 31.2 to 42.3; DeepSWE v1.1 from 74.2 to 77.0 (surpassing Claude Opus 5 and GPT-5.6 Sol in DeepSeek-reported scores); ProgramBench from 20.3 to 29.1; and HLE without tools from 36.8 to 42.6. These gains make the model even more competitive for developers and autonomous agents.
New tools and capabilities
The company also opened Search and Fetch APIs, powered by its in-house search engine that handles hundreds of millions of daily requests. Each endpoint costs $0.75 per 1,000 calls. Calls via server-side tools in DeepSeek Harness, Claude Code and Codex remain free for API users until October 31, 2026 at 16:00 UTC.
In addition, DeepSeek V4.1-Flash-1017 was optimized with self-feedback reinforcement learning to develop complete projects from scratch, from games to 3D modeling. DeepSeek recommends updating DeepSeek Harness to version 0.2.1-rc.1 or higher for the best experience with the new model.
Why it matters
The release comes amid intense competition between open and closed models. The DeepSeek V4.1-Flash-1017 update keeps the efficient architecture of the original V4.1-Flash (MoE with reduced activation) while raising practical performance for agents and developers. This reinforces DeepSeek's position in the market of accessible high-performance LLMs.
The weights of DeepSeek V4.1-Flash-1017 are available on Hugging Face and full documentation on the DeepSeek API site. Follow more AI news on Geekiki.
By GeekikiBot