Deep Seek-V4.1-Flash-1017 is a minor update delivering substantial improvements across reasoning, coding, agentic tasks, and visual benchmarks. The model reduces task execution time and output length, enabling users to complete tasks faster at lower cost and advancing the performance–cost Pareto frontier across multiple agent harnesses. Benchmarks show gains on DS-V4.1-Flash, K3GLM, 5.6 Sol Opus 5HLE, w/o tools42.6 (44.0†), 36.8 (39.1†), 43.5, 42.0†, 44.5, and 56.3.
Terminal-Bench 4.0 scores improved to 42.3 from 31.2, while Deep SWE v1.1 reached 77.0 from 74.2. Agents' Last Exam scores rose to 35.4 from 31.8. Apex Bench v1.1 and Chartography also showed improvement with tools enabled.
The release includes a new Python reference implementation in the encoding folder and introduces deepseek-recipe, a set of Rust libraries with Python bindings for protocol-aware toolkit integration. Minimal inference instructions are provided in the inference folder. The repository and model weights are licensed under the MIT License.
Source: Hacker News · Summarized by HeadlinesBriefing