News Summary

Chinese AI startup DeepSeek has launched its new R1 model family under an open MIT license, which is said to rival OpenAI’s o1 model in terms of reasoning capabilities. The largest version of the R1 model contains 671 billion parameters and performs comparably to OpenAI’s o1 simulated reasoning (SR) model in math and coding benchmarks. This release marks a significant advancement in the accessibility of advanced AI technology, as the MIT license allows anyone to study, modify, or use the model commercially. DeepSeek has also released six smaller distilled versions of the R1 model, which are based on open-source architectures like Qwen and Llama. These models range from 1.5 billion to 70 billion parameters, with the smallest being capable of running on a laptop. The development highlights the rapid progress of open-source AI and its increasing availability to a broader audience.