DeepSeek-R1

January 25, 2025

DeepSeek-R1 has surpassed O1 from OpenAI, which had previously been the leading state-of-the-art foundation model by a wide margin. Despite this, DeepSeek-R1 had far fewer parameters, making it cheaper and capable of running on consumer computer hardware. However, I was skeptical about the numbers they achieved on the datasets, as most of them were likely contaminated to some extent