DeepSeek updates its V4.1-Flash model and opens its in-house search engine to developers
China’s DeepSeek released V4.1-Flash-1017 with open weights under the MIT licence, claiming stronger coding and agent scores with shorter outputs, and opened Search and Fetch APIs built on its own search engine at $0.75 per 1,000 calls.
This story was prepared automatically by software.
Chinese AI developer DeepSeek on October 17 released DeepSeek-V4.1-Flash-1017, an update to its V4.1-Flash model that it says delivers “stronger benchmark performance, faster task completion, and shorter outputs than the previous version — a Pareto improvement in performance and cost.” The model is live on DeepSeek’s website, app and developer platform, Bloomberg reported.
What changed
DeepSeek described the release as a minor update to V4.1-Flash, according to Bloomberg, while its technical report on Hugging Face cites “substantial improvements across reasoning, coding, agentic tasks, and visual benchmarks.” The company said it used reinforcement learning with self-feedback to improve end-to-end project development, showcasing examples ranging from a 3D model of Beijing’s Forbidden City to a web-based racing game.
On DeepSeek’s published tables, the update scores 77.0 on DeepSWE v1.1 (up from 74.2), 42.3 on Terminal-Bench 4.0 (from 31.2) and 42.6 on Humanity’s Last Exam without tools (from 36.8). DeepSeek puts it ahead of Anthropic’s Claude Opus 5 (74.0) and OpenAI’s GPT-5.6 Sol (73.0) on DeepSWE, though Opus 5 still leads on Terminal-Bench 4.0 and Humanity’s Last Exam. All figures are vendor-reported and have not been independently verified.
The weights are available on Hugging Face and ModelScope under the MIT licence, with a context window of 1 million tokens.
Search as a product
Alongside the model, DeepSeek opened Search and Fetch APIs to all developers, built on an in-house search engine that it says serves hundreds of millions of requests a day. Each endpoint costs $0.75 per 1,000 successful calls. Calls made through server-side tools in DeepSeek Harness, Claude Code and Codex are free for DeepSeek API users until billing begins on October 31 at 16:00 UTC. The move means DeepSeek is now selling a retrieval layer for AI agents, not just models.
Competitive backdrop
Bloomberg said the rapid pace of DeepSeek’s launches underscores the intensifying contest between China’s open-weight models and the proprietary models of US rivals such as Anthropic and OpenAI, which are top performers but far more expensive. DeepSeek also competes with domestic players including Moonshot’s Kimi and Z.ai.
Earlier in the week DeepSeek suffered an outage of about five hours, which some experts attributed to a lack of computing resources, and slowdowns have increased in recent months as demand for its aggressively priced models grows. When DeepSeek launched V4.1-Flash in early September at a steep discount, shares of rivals including MiniMax, Z.ai and Alibaba fell sharply in Hong Kong, Bloomberg noted.
Sources: Bloomberg (via Yahoo Finance), DeepSeek API news, Hugging Face model card. Image: DeepSeek.
Related articles
