DeepSeek V4.1 Flash: Why Developers Are Leaving Claude
Something interesting happened in AI developer circles this weekend. On October 17, DeepSeek pushed out a quiet update to its cheapest coding model — V4.1-Flash-1017 — and within hours developers on Hacker News were talking about switching to it from Claude and Gemini. DeepSeek calls it "a minor update" — faster task completion, shorter outputs. But in AI coding tools, the model developers actually enjoy using often beats the one that wins on paper. What DeepSeek Shipped on October 17 V4.1-Flash-1017 is a refresh of the V4.1-Flash model from early September. It went live the same day across DeepSeek's website, app, and API — and if you already call the deepseek-flash endpoint, you got the upgrade automatically. Old endpoint names still work, served by the new weights at the same price. The weights are also on Hugging Face and ModelScope under an MIT license, so commercial use and redistribution are allowed. The model keeps its 1-million-token context window with tex...