DeepSeek Releases V4.1-Flash-1017, MIT Open-Source Weights Launched in Tandem

DeepSeek released V4.1-Flash-1017 on October 17, a minor-version update to V4.1 Flash that focuses on being faster, lighter and cheaper. The new model has 552B total parameters (8B input activation, 16B output activation), a 1-million-token context window, and its weights are open-sourced on Hugging Face under the MIT license. The company says the new version delivers across-the-board improvements on reasoning, coding and agentic tasks, with a DeepSWE v1.1 score of 77.0, up from 74.2 for the previous generation; API pricing remains unchanged at $0.15/$0.60 per million tokens, and calls still go through the deepseek-flash alias. DeepSeek also launched Search and Fetch APIs in the same period, charging $0.75 per thousand successful requests.