Skip to content

DeepSeek ships V4.1-Flash-1017, open weights under MIT

A minor-numbered update to V4.1-Flash that DeepSeek says improves coding, reasoning and agentic benchmarks while cutting run time and cost, weights released on Hugging Face.

Source: DeepSeek

DeepSeek released V4.1-Flash-1017 on 17 October, an update to V4.1-Flash that it calls a “Pareto improvement”: better benchmark scores, faster task completion and shorter outputs than the version it replaces. It is live now on DeepSeek’s website, app and API, billed as deepseek-flash, and the weights are on Hugging Face under an MIT licence.

The model card lists 763 billion total parameters and a 1M-token context window. DeepSeek’s own numbers show the jump on DeepSWE v1.1, its agentic coding benchmark: 77.0 for the new release against 74.2 for the previous V4.1-Flash. The company says the gains carry across reasoning, coding, agent tasks and image understanding.

Why it matters: at 763B parameters, self-hosting this needs serious GPU memory, but the MIT licence means anyone can fine-tune or redistribute it without the restrictions attached to some open-weight releases. It is also a live benchmark for what a frontier Chinese lab considers “minor” progress on agentic coding right now.

The caveat: the benchmark numbers are DeepSeek’s own, measured against its own prior model. There is no independent comparison yet against other labs’ current models.

Share

Related reading

ML for Web Devs

Tokens, context windows, and why your bill is what it is

The unit of billing for language models is not the request or the word. Once you understand tokens and how a context window fills up, the invoice stops being a surprise.

ML for Web Devs

RAG is just search with extra steps

Retrieval-augmented generation gets written up like an architecture. It is a search query, a string concatenation and one API call, and the search query is the part that decides whether it works.

ML for Web Devs

A prompt is a spec, so write it like one

Prompt engineering has very little to do with magic words. It is the same skill as writing a clear ticket for a capable contractor who will not ask you any questions.

More from The Wire

All briefs

GitHub Copilot retires six AI models today

Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini and Grok 4.5 stop working across Copilot Chat, inline edits, agent mode and completions from today.

via GitHub