AI has been transforming different spheres in the last three years. AI's journey in mathematics has transformed dramatically, evolving from failures in basic arithmetic to elite competition dominance. Nous Research's open-source Nomos-1, a 30B parameter model, scored 87 out of 120 on the 2025 Putnam Contest — a score placing second among nearly 4,000 human competitors last year, with eight perfect problems.
This two-phase system uses AI workers for self-critiquing solutions in a tournament bracket, far surpassing Qwen3's 24/120 under identical conditions. This breakthrough joins 2025's wave of AI math advances, including OpenAI's gold medal-level performance at the International Mathematical Olympiad (35/42 points on six problems) and AxiomProver autonomously solving 9/12 Putnam problems.
Other milestones feature DeepMind's Gemini Deep Think and grants accelerating AI theorem proving via tools like Lean4. These feats highlight specialized training and reasoning harnesses over raw scale. The trajectory points to an AI-driven boom in mathematical discovery, with multi-agent systems, symbolic reasoning integration, and dynamic prompts enhancing problem-solving.
Mathematics is a fascinating field, where AI enabled seminal research can be highly utilitarian. Expect AI to tackle unsolved problems, formalize proofs, and support education through personalized tutoring and code verification. Open-source efforts like Nomos democratize access, challenging parameter-size dominance.
AI IS POISED TO UNLOCK MATHEMATICS' DEEPEST SECRETS!
Sanjay Sahay
Have a nice evening,