Skip to content- Sources: primary, discussion, discussion
- Summary: DeepSeek-V4.1-Flash-1017 is a minor update to V4.1 Flash that DeepSeek says improves reasoning, coding, agentic and visual benchmarks while cutting execution time and output length. The Hugging Face repository lists 763B parameters in FP8 and I8 tensors, a 1M-token context window, and an MIT license. It ships no Jinja chat template, and instead provides a Python reference encoder and deepseek-recipe, Rust libraries with Python bindings for prompt encoding and output parsing.
- Why it matters: Teams serving DeepSeek locally must adopt the new encoder or deepseek-recipe instead of a chat template, and get a permissively licensed model with vendor-reported Terminal-Bench gains.