AI Models Rapidly Advancing in Theoretical Physics and Economics, MIT Study Finds

Frontier artificial intelligence models are expanding beyond pure mathematics into a broad spectrum of theoretical sciences, according to findings from researchers at the Massachusetts Institute of Technology first reported by Wired (https://www.wired.com/story/ai-is-coming-for-science-next-watch-out/). A new study examining the capability trajectory of frontier AI systems indicates that domains such as quantum physics, economic theory, and gravitation and cosmology may experience the same disruption currently reshaping mathematical research.

The study was conducted by Neil Thompson, who leads research on computing progress at MIT, alongside doctoral student Gabriel Manso and research scientist Claudio Mayrink Verdun. Their analysis found that between mid-2024 and mid-2026, the success rate of leading AI models on theoretical problems across several scientific fields jumped from approximately 2 percent to about 35 percent. The research evaluated performance across probability, statistics theory, economic theory, high-energy theory, gravitation and cosmology, and quantum physics.

“Within just a few years, we expect a lot of theory questions to be totally transformed in terms of AI's ability to do them,” Thompson said during a presentation at MIT. The researchers focused specifically on theoretical disciplines where models could derive solutions without conducting physical experiments or running computer simulations.

To benchmark capabilities without data contamination, the MIT team selected theoretical problems from research papers posted online between December 2025 and May 2026. The researchers evaluated OpenAI models—including o1, o3, GPT-5, GPT-5.1, GPT-5.4, and GPT-5.5—with internet access disabled. Because the target papers were published after the training cutoff for GPT-5.5 and preceding models, the systems had to generate original solutions rather than recall existing text.

The findings follow growing debate over automated discovery in the scientific community. Last month, OpenAI announced that an unreleased model had solved a long-standing challenge concerning the Navier-Stokes equations governing fluid motion. The announcement sparked controversy after a mathematician at New York University, collaborating with an academic associated with Anthropic, accused OpenAI of gazumping them after learning of their work. Since then, OpenAI and others have announced hundreds more solutions to mathematical problems, prompting some mathematicians to call for limits on AI usage in research.

Despite rapid capability gains, the MIT researchers emphasized that human domain experts remain essential for framing research questions and validating model reasoning. Theoretical models can produce convincing yet flawed answers; OpenAI itself recently withdrew several problem solutions from its latest dataset post-publication. “We are still going to need people to generate good ideas to connect the dots,” Manso said. “We also need ways to verify what was done, because it's very easy for a model to tell you ‘this is the right answer,’ and when you look at the entire solution, you start to see maybe it is not.”