AnalysisScienceOctober 14, 2026

MIT study: AI models jumped from 2% to 35% on theory problems

Read original source →wired.com

MIT researchers tested OpenAI's o1, o3, GPT-5, GPT-5.1, GPT-5.4 and GPT-5.5 on theory problems from papers published after the models' training cutoff, with internet access blocked. Correct-solve rates rose from about 2% to about 35% between mid-2024 and mid-2026 across probability, statistics, economic theory, high-energy theory, gravitation and cosmology, and quantum physics.

People · Neil Thompson

1 source

More stories today

Open the live feed