LMarena rankings published at the end of 2025

DeepSeek-R1 Matches OpenAI's o1, 20x Cheaper and Open-Source

Jan 24, 2025, 09:49 PM

Chinese AI startup DeepSeek has released an open-source AI model, DeepSeek-R1, that rivals leading models from U.S. tech giants such as OpenAI and Google while being significantly more cost-effective and efficient. The model matches the performance of OpenAI's o1 on various benchmarks, including ranking #3 overall on the LMarena rankings and securing #1 in technical domains like coding, mathematics, and hard prompts. Remarkably, DeepSeek-R1 is 20 times cheaper and five times faster than OpenAI's o1, and it is MIT-licensed, emphasizing the company's commitment to openness. Despite U.S. export controls limiting access to high-end chips, DeepSeek developed the model using 2,048 NVIDIA H800 GPUs, completing training in just two months at a cost of $5.5 million. This achievement has raised concerns about the effectiveness of U.S. trade restrictions and the massive capital investments made by American tech companies in AI development. Industry observers note that DeepSeek's success underscores China's rapid progress in AI, leveraging efficiency and collaboration to challenge the dominance of U.S. firms and potentially reshaping the balance in the global AI race.