DeepSeek’s open-source R1 beats OpenAI o1

Chinese AI lab DeepSeek just released DeepSeek-R1, an open-source reasoning model that reportedly matches or exceeds OpenAI's o1 on certain benchmarks while costing just 5-10% of o1's API price for developers.

Unlike traditional GPT models, R1 uses a reasoning approach similar to OpenAI’s o1 that takes longer but produces more reliable results in domains like physics, science, and math.

The model contains 671B parameters but also comes in smaller "distilled" versions, with as few as 1.5B parameters, that can run locally on a laptop.

DeepSeek-R1 beats o1 on several key benchmarks, including AIME, MATH-500, and SWE-bench Verified.

The model is available under an MIT license for commercial use and costs significantly less than o1 ($0.14 vs $7.5 per million input tokens).

Why it matters: Open-source AI just achieved a significant milestone by matching ChatGPT's current capabilities on key benchmarks. And in an ironic twist, it's not OpenAI (which abandoned its original mission of open-source research) but Chinese company DeepSeek, openly sharing its models and training methodology.