DeepSeek Just Released V4.1-Flash-1017, Scores 77.0 on DeepSWE, Opens Search API

DeepSeek has pushed out a new checkpoint of its V4.1-Flash model, called DeepSeek-V4.1-Flash-1017. It’s live on the company’s website, app, and API, and DeepSeek says the update is better on performance, cost, speed, and output length than the September release. That’s the pitch: the same model family, tightened up, with less waste.

The headline number is DeepSWE v1.1, where DeepSeek says the model scores 77.0, up from 74.2 on the earlier V4.1-Flash. The company says the new checkpoint beats the previous one on all 11 benchmarks it reported, by between 2.0 and 11.1 points. It also claims the model leads on Automation-Bench v1.0 and Agents’ Last Exam, while Claude Opus 5 still comes out ahead on Terminal-Bench 4.0, ProgramBench, NL2Repo-Bench and HLE without tools.

The architecture is not a small one. DeepSeek lists 552B backbone parameters plus 196B Engram parameters, with 8B active during prefill and 16B during decode. Context goes up to 1M tokens, and output can reach 384K tokens. The older September V4.1-Flash weights were released on Hugging Face under MIT, but the source says hardware requirements for 1017 and its weights were not disclosed.

There’s a second story here too. DeepSeek has opened its Search and Fetch APIs to every developer, so agent builders can pair a coding model with a first-party web layer behind one key. Search returns ranked web results, Fetch pulls a single page as Markdown, and both cost $0.75 per 1,000 successful requests. Calls through DeepSeek Harness, Claude Code, or Codex stay free until November 1, 2026.

DeepSeek is also leaning hard into agent use. The company says the checkpoint uses reinforcement learning with self-feedback for end-to-end project work, with examples ranging from game creation to 3D modeling. It is tuned for DeepSeek Harness, Tencent CodeBuddy, and WorkBuddy, and DeepSeek recommends Harness v0.2.1-rc.1 or later. The message is pretty clear: this is not just another model drop, it’s an attempt to bundle search, fetching, and coding into one cheaper pipeline.