DeepSeek launches DeepSeek‑V4.1‑Flash‑1017 model with higher benchmark scores

Desmond Otoo

Oct 18, 2026 at 05:02 AM Updated: Oct 18, 2026 at 05:02 AM

DeepSeek released its DeepSeek‑V4.1‑Flash‑1017 model, citing faster performance and lower cost.

On Saturday, Chinese artificial‑intelligence startup DeepSeek announced that its updated model, DeepSeek‑V4.1‑Flash‑1017, is now available through the company’s website, mobile application and API.

The firm described the release as a “Pareto improvement” in performance and cost, indicating that the model delivers greater capability without an increase in expense.

Benchmark testing showed the model achieved a score of 77.0 on the DeepSWE v1.1 benchmark, up from 74.2 for the previous version, and a score of 60.5 on Automation‑Bench V1.0.6, up from 54.8, according to data shared by DeepSeek.

DeepSeek‑V4.1‑Flash‑1017 has been optimised for DeepSeek Harness, Tencent CodeBuddy and WorkBuddy, among other tools. The update also incorporates reinforcement learning with self‑feedback to improve end‑to‑end project development, including game creation and the generation of 3D models from scratch.

Search and Fetch APIs opened to all users

In parallel with the model launch, DeepSeek opened its Search and Fetch API services to all users. Each endpoint is priced at US$0.75 per 1,000 calls and is powered by the company’s in‑house search engine.