AI Models Solve 35% of Theoretical Physics and Economics Problems
An MIT study found that top AI models improved from solving about 2% to 35% of theoretical problems in physics, economics and statistics between mid-2024 and mid-2026. OpenAI's o1, o3 and GPT-5.x models were tested on problems from papers published after their training cutoff, with no internet access.
- Theoretical problem-solving rose from 2% to 35% in two years
- o1, o3, GPT-5, GPT-5.1, GPT-5.4 and GPT-5.5 tested offline
- Problems came from papers published December 2025 to May 2026
- Fields include probability, economics, cosmology and quantum physics
Read next
AI