DeepSeek updates V4.1 Flash and opens its search engine to developers

Chinese AI lab DeepSeek released DeepSeek-V4.1-Flash-1017 on October 17, 2026, an update to its Flash model that is live in its app, website and API and also published as open weights on Hugging Face and ModelScope under the MIT license. The Decoder covered the launch the same day.

DeepSeek calls it a minor update but says it scores higher on every benchmark it reported while finishing tasks faster and writing shorter answers. In the company's own tests it beats larger models such as Claude Opus 5 and GPT-5.6 Sol on several coding and agent benchmarks, including 77.0 on DeepSWE. DeepSeek says that result costs just under 20 cents per task, against about 50 cents for the previous version. These are DeepSeek's own numbers, and independent results have not appeared yet. The company credits a training method in which the model uses its own confidence and critiques as a reward signal.

DeepSeek is also opening its in-house web search to outside developers through new Search and Fetch APIs, priced at $0.75 per 1,000 calls. API users can try them free inside some coding tools until October 31.

Why it matters

A cheap, open-weight model that is strong at agent work gives teams a real option to cut the cost of coding agents or run them on their own hardware.

Source: The Decoder