DeepSeek Open-Sources IMO Gold-Medal Math Model Math-V2
DeepSeek-Math-V2: open-source IMO-gold prover that self-checks every step.
“AI Disruption” Publication 8300 Subscriptions 20% Discount Offer Link.
Breakthrough-Level Reasoning Model Arrives: DeepSeek Opens Up the Direction of Self-Verifiable Mathematical Reasoning.
The whale is back!
Just moments ago, DeepSeek quietly uploaded a new model to Hugging Face: DeepSeek-Math-V2.
As the name suggests, this is a mathematics-focused model. Its previous version—DeepSeek-Math-7b—was released over a year ago. At that time, the model achieved performance comparable to GPT-4 and Gemini-Ultra using only 7B parameters. The related paper also introduced GRPO for the first time, significantly enhancing mathematical reasoning capabilities.
So after a year and a half, what surprises does DeepSeek-Math-V2, developed based on DeepSeek-V3.2-Exp-Base, bring?
DeepSeek states that its performance surpasses Gemini DeepThink, achieving IMO gold medal-level capabilities.