⚡️ DeepSeek drops Math-V2 and it’s a monster
No hype, no teasers, DeepSeek just pushed out Math-V2 almost two years after the original. No one was waiting for it… and yet it quietly arrived with absurd numbers.
Performance snapshot:
• Gold-medal strength on IMO 2025 and China’s CMO 2024.
• 118/120 on Putnam 2024 , a famously brutal university-level contest.
• Sparse benchmarks (no MATH, GSM8K, AIME), but IMO-ProofBench shows it outscoring GPT-5 and matching Gemini 2.5 Pro on hard proofs. Now everyone’s watching what Gemini 3, Grok 4, and GPT-5.1 will do next.
What’s inside the system:
• Bui
No hype, no teasers, DeepSeek just pushed out Math-V2 almost two years after the original. No one was waiting for it… and yet it quietly arrived with absurd numbers.
Performance snapshot:
• Gold-medal strength on IMO 2025 and China’s CMO 2024.
• 118/120 on Putnam 2024 , a famously brutal university-level contest.
• Sparse benchmarks (no MATH, GSM8K, AIME), but IMO-ProofBench shows it outscoring GPT-5 and matching Gemini 2.5 Pro on hard proofs. Now everyone’s watching what Gemini 3, Grok 4, and GPT-5.1 will do next.
What’s inside the system:
• Bui
