DeepSeek Updates V4.1‑Flash AI Model

The company described the update as a “minor upgrade” that reduces task execution time, and the new version is already available on DeepSeek’s website, mobile app, and API.

Performance Gains

The company also highlighted broad improvements in reasoning, coding, agentic tasks and visual challenges, and said the update boosts video‑game creation and 3D modelling through reinforcement learning with self‑feedback.

Pricing for the model remains unchanged, with off‑peak API rates as low as 15 cents per million uncached input tokens. DeepSeek also added two new APIs – a Search API that returns up to 50 ranked web results and a Fetch API that converts a single page to Markdown – both priced at 75 cents per 1,000 successful requests. Calls made through DeepSeek Harness, Anthropic’s Claude Code or OpenAI’s Codex are free until November 1, and the company offered credits to users who suffered from an outage on October 15.

The release sent shares of DeepSeek rivals MiniMax Group and Alibaba down in Hong Kong, as Bloomberg noted that the launch of a competitively priced open‑weight model had shaken investor confidence.

DeepSeek also tuned the update for agent software, supporting its own Harness tool and two Tencent products – the CodeBuddy coding assistant and WorkBuddy, a desktop office agent – and has opened its search engine to external developers. The updated weights are freely available on Hugging Face under an MIT licence.