DeepSeek-V4.1-Flash-1017 opens MIT-licensed weights

DeepSeek released V4.1-Flash-1017 through its official Hugging Face organization as a minor update to V4.1-Flash. The model card reports substantial gains across reasoning, coding, agentic and visual benchmarks while reducing execution time and output length. Its published comparison moves Terminal-Bench 4.0 from 31.2 to 42.3, DeepSWE v1.1 from 74.2 to 77.0 and Automation-Bench from 54.8 to 60.5. The image-text model supports a one-million-token context, at least 256,000 output tokens and documented use through Transformers, vLLM, SGLang and Docker. Weights use the MIT license, but inference, tool execution and HTTP transport remain the caller's responsibility, so real cost, safety and long-task reliability still require independent evaluation.