DeepSeek, a China-based AI lab, has unveiled its groundbreaking open-weight model, DeepSeekMath-V2, showcasing exceptional theorem-proving abilities. This model achieved gold-level performance at the International Mathematics Olympiad (IMO) 2025, solving 5 out of 6 problems. Hugging Face’s CEO, Clement Delangue, emphasized the model’s uniqueness, noting its availability as a free resource akin to having the mind of a top mathematician. Alongside DeepSeek, Google DeepMind’s Gemini and OpenAI also reached gold status at IMO 2025. In addition to IMO, DeepSeekMath-V2 demonstrated stellar results in the China Mathematical Olympiad and the Putnam exam, where it scored 118 out of 120, surpassing the highest human score. DeepSeek’s innovative approach trains a separate verifier to assess proofs critically, reinforcing self-verification for enhanced reasoning, crucial for tackling mathematical challenges. The model is available for download on Hugging Face, representing a significant step in the democratization of AI and mathematical knowledge.
Source link
