𝗠𝗶𝗻𝗶𝗠𝗮𝘅 𝗠𝟯 𝗕𝗲𝗮𝘁𝘀 𝗛𝘂𝗺𝗮𝗻 𝗠𝗮𝘁𝗵 𝗚𝗼𝗹𝗱 𝗠𝗲𝗱𝗮𝗹𝘀
MiniMax claims its new M3 model outperformed human gold medalists on math benchmarks.
The company uses a new method called MaxProof to reach these results. This framework helps the model reason through math problems step by step.
You should note a few things about this news:
- MiniMax did not share specific math scores.
- The names of the benchmarks are not public.
- The full research paper is not yet available for review.
Exceeding a gold medal threshold usually means a model performs well on difficult exams like AIME or AMC. This requires high levels of logical reasoning.
MaxProof likely works by verifying proofs to stop the model from making mistakes. This addresses a common problem in AI where models guess wrong numbers.
We do not know if M3 wins through more data, better design, or new training methods. Until researchers see the full paper, these results remain unverified.
This move shows MiniMax is moving into the reasoning market. They are now competing with models from OpenAI and Anthropic.
Watch for the full paper to see if independent testers can repeat these results.
Source: https://dev.to/gentic_news/minimax-m3-exceeds-human-gold-medal-on-math-benchmarks-via-maxproof-4pph
Optional learning community: https://t.me/GyaanSetuAi