MIT study shows AI solving more theoretical science problems by mid-2026
Neil Thompson of MIT presented a study last week showing that frontier AI models solved about 35 percent of theoretical problems in fields like physics and economics between mid-2024 and mid-2026.
Study covers multiple scientific fields
The research team tested OpenAI models against problems found in papers posted online from December 2025 to May 2026. They included probability theory, economic theory, high-energy physics, gravitation, cosmology, and quantum mechanics in their analysis. Solutions were identified even after the training cutoff for the GPT-5.5 model.
Performance increased significantly over time
Thompson noted that the ability of these models to solve theoretical questions rose from 2 percent to 35 percent during the study period. The researchers expect this progress in AI capability to continue without slowing down soon. This transformation affects how theory questions are approached across several disciplines.
Reported by one outlet
Only one outlet has published this. Nothing here has been checked against a second report, so read it as that outlet's account and follow the link below for the original.
Reported by
1 independent outlet. Headline as published. Links open the original report.