DeepSeek’s October 17 changelog confirms that V4.1 Flash was upgraded to DeepSeek-V4.1-Flash-1017 across its web service, app and API. API callers continue to use deepseek-flash, with the calling method and prices unchanged.
DeepSeek reports improvements in reasoning, coding, agent tasks and visual benchmarks, alongside shorter outputs and reduced task runtime. The published benchmark numbers are the company’s evaluations, not a blanket claim that it leads every model.
The model card provides MIT-licensed weights and inference references. DeepSeek also recommends Harness 0.2.1-rc.1 or later for this update; that harness remains separate from the model itself.
Sources: Official model card · October 17 changelog
