DeepSeek-V4.1-Flash-1017: DeepSeek's minor update, open under MIT

DeepSeek has released DeepSeek-V4.1-Flash-1017 on Hugging Face under an MIT license. It is a minor update to DeepSeek-V4.1-Flash and handles both text generation and image-text-to-text tasks.

DeepSeek says the update brings substantial improvements across reasoning, coding, agentic tasks and visual benchmarks. It also says the model cuts task execution time and output length, so tasks finish faster and at lower cost. In DeepSeek's own table, it scores 77.0 on DeepSWE v1.1 against 74.2 for V4.1-Flash, and 42.3 on Terminal-Bench 4.0 against 31.2.

The weights are on Hugging Face in fp8 and 8-bit formats and work with the transformers library. The release has no Jinja chat template. Instead it ships a Python reference encoder and deepseek-recipe, a set of Rust libraries with Python bindings for production prompt encoding and output parsing. DeepSeek recommends temperature 1.0 and top_p 1.0, with a 1M token context window and max_tokens of at least 256K.

The model builds directly on DeepSeek-V4.1-Flash, which the table lists as the baseline. The same table compares it with other models, including K3, GLM-5.3, GPT-5.6 Sol and Opus 5, and all of those figures are DeepSeek's own.