Deep Noodle

Deep Noodle

AboutBlogAtlas
Get in touch

DeepSeekMath-V2: 118/120 Putnam Score and IMO Gold-Level Result

DeepSeek released its 685-billion-parameter Math-V2 under Apache 2.0 and reported a 118/120 Putnam score plus an IMO 2025 gold-level result.

DeepSeek

Sources

arXiv

Paper · Nov 27, 2025

DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning

Additional sources

Simon Willison

Analysis · Nov 27, 2025

deepseek-ai/DeepSeek-Math-V2

previous event
Zig moves to Codeberg amid GitHub reliability and AI criticism
next event
DeepSeek puts gold-medal reasoning in MIT-licensed weights