DeepSeek ships V4.1-Flash-1017 for building whole projects from scratch, and opens its search and fetch APIs to everyone
Synopsis
DeepSeek released DeepSeek-V4.1-Flash-1017, which it says delivers stronger benchmark performance, faster task completion and shorter outputs than the previous version, a Pareto improvement in performance and cost, now live on its website, app and API. The model uses reinforcement learning with self-feedback to improve end-to-end project development, with demos spanning 3D modeling, a web racing game and a Godot-engine recreation. The Search and Fetch APIs also opened to everyone at $0.75 per 1,000 calls per endpoint.
Interpretation
DeepSeek has put DeepSeek-V4.1-Flash-1017 into service on its website, app and API, and published the model weights on Hugging Face and ModelScope. The release compares it with the previous version, reporting stronger benchmark performance, faster task completion and shorter outputs, and describes that combination as a Pareto improvement in performance and cost. The comparison rests on the release note itself, which names no benchmarks, scores or cost figures, so comparability currently stays at the level of the publisher's statement.
The model targets building complete projects from scratch, with demos covering 3D modeling of the Forbidden City, a web-based racing game, a recreation built with the Godot engine and a 3D scene tracing prehistory to the modern era. The stated method is reinforcement learning with self-feedback, meaning the model adjusts its behavior during training using feedback signals it produces itself, applied to end-to-end project development. These appear as demo entries in the release note, indicating the task types covered; no completion quality or evaluation result is reported for them.
The Search and Fetch APIs are now open to everyone, powered by DeepSeek's in-house search engine, which the release says serves hundreds of millions of requests daily worldwide. Pricing is $0.75 per 1,000 calls per endpoint, and calls made through server-side tools in DeepSeek Harness, Claude Code and Codex are currently free for DeepSeek API users. Billing starts Oct 31, 2026, at 16:00 UTC, a date the release note states explicitly.
Perspective
This report covers DeepSeek's release note: the new model is live on the website, app and API, weights are public on Hugging Face and ModelScope, and the Search and Fetch APIs are open at $0.75 per 1,000 calls. The note recommends upgrading DeepSeek Harness to v0.2.1-rc.1 or later for the best experience and lists coding tools the model is optimized for.
The claimed stronger benchmark performance, faster task completion and shorter outputs come with no benchmarks or numbers in the release note. The Pareto improvement is the publisher's description of the performance-and-cost combination and lacks checkable cost data. The demos list task types only and report no completion quality. The free search and fetch arrangement has a stated cutoff, after which list pricing applies.