DeepSeek has officially retired its R1 reasoning model, marking the end of an era for the open-source AI community. The company announced that the deepseek-reasoner alias was decommissioned on July 24, 2026, and all new development should now target the DeepSeek V4 model.
R1, which launched in January 2025, was a pioneering open-source reasoning model known for its chain-of-thought capabilities, strong coding performance, and state-of-the-art mathematical reasoning. It was released under the MIT license, allowing free download and self-hosting.
While R1's weights remain available for local use, DeepSeek advises users to migrate to V4 for current API access, pricing, and benchmarks. The company provides a migration guide and a free Chrome sidebar to ease the transition.
For users still running R1 on their own hardware, the model continues to function, but DeepSeek recommends focusing new work on V4 to benefit from the latest advancements.