DeepSeek releases V4.1 Flash

DeepSeek describes V4.1-Flash-1017 as a minor update to V4.1-Flash, with gains in reasoning, coding, agent tasks and visual benchmarks. The company also says the model completes tasks in less time and produces shorter outputs, which could lower costs in some uses. Those are the company’s claims, so businesses should test the model on their own work before assuming the same gains will show up in practice. The model page and technical report include benchmark results and setup details.

Some of the reported improvements are notable: Terminal-Bench 4.0 rises from 31.2 to 42.3, and ProgramBench from 20.3 to 29.1, compared with the previous Flash model. The page also reports stronger scores on visual benchmarks, including Chartography and BabyVision. Benchmark scores can help narrow down what to test, but they don’t establish how well a model will handle your company’s specific tasks.

For teams considering a trial, the repository includes prompt-encoding tools and instructions for running inference locally, and the model is licensed under MIT. This is the kind of thing we build into an AI command centre, where teams can track models and decide what’s worth testing against real business processes.