Cutting-edge Chinese “reasoning” model rivals OpenAI o1—and it’s free to download
1 min read
Summary
Chinese AI lab DeepSeek has attracted attention from the wider AI community by releasing its R1 model family, which the company claims performs at the same level as OpenAI’s simulated reasoning (SR) model on math and coding benchmarks, under an open MIT licence.
The R1 models require substantial computing resources, but the firm also published six smaller ‘distilled’ versions of the model based on existing open source architectures, which range from 1.5 billion to 70 billion parameters and can run on a laptop.
The release reflects a shift in what is possible with open AI models, and demonstrates China’s increasingly powerful role in the AI space.
These models employ an inference-time reasoning approach to simulate human-like chains of thought, and emerged when OpenAI released its o1 model in 2024.
The AI community has been buzzing about DeepSeek owing to the cost efficiency of its model’s training and running processes compared with OpenAI.
