DeepSeek
Open-source LLM optimized for advanced reasoning and code
Start new thread
Zac Zuo

DeepSeekMath-V2 - IMO Gold level reasoning, fully open.

by
Flowtica Scribe
•
DeepSeekMath-V2 is a new open-source model specialized in mathematical reasoning. It introduces a self-verification mechanism where the model acts as both generator and verifier to refine its own proofs. It achieved Gold-level scores in IMO 2025 and a near-perfect 118/120 in Putnam 2024.

Add a comment

Replies

Best
Zac Zuo
Flowtica Scribe
Hunter
📌

Hi everyone!

DeepSeekMath-V2 is DeepSeek's latest released math model, and the results are wild. It scored 118/120 on Putnam 2024, beating the top human score of 90...🤫

Putnam Competition is extremely hard. According to Wiki:

Each of the twelve questions is worth 10 points......The competition is considered to be very difficult: it is typically attempted by students specializing in mathematics, but the median score is usually zero or one point out of 120 possible, and there have been only five perfect scores as of 2021.

So the score it achieved is astonishing and the core logic here is fascinating: it reaches a higher ceiling not by just guessing the answer, but through a process of "self-distrust" or self-challenge. It rigorously checks its own reasoning steps to ensure the logic holds up.

In the landscape of IMO Gold-level models, OpenAI's version isn't officially out yet (maybe part of it is in GPT-5?) Google has Gemini 2.5 Deep Think, but the public version is a variant. DeepSeekMath-V2 is performing at that same elite level, and it's perhaps the only one people can actually generally access today.