Release anniversaries, 20 Jan 2027
DeepSeek-R1 release
On 20 January 2025 DeepSeek released R1, an open reasoning model under the MIT licence.
DeepSeek released R1 on 20 January 2025, with weights under the permissive MIT licence and a paper explaining its training. It matched OpenAI's o1 on many reported maths and coding benchmarks. The paper showed that a variant, R1-Zero, learned long step-by-step reasoning through reinforcement learning alone. DeepSeek also released smaller models distilled from R1 onto Qwen and Llama bases. Within a week its app topped download charts in the United States, prompting wide debate on export controls and the cost of AI. The paper was later published in Nature.