TheoPhysEval: Assessing AI agents at autonomous theoretical physics problem solving

@misc{chandra2026theophyseval,
  title = {{TheoPhysEval: Assessing AI agents at autonomous theoretical physics problem solving}},
  author = {A. Ramesh Chandra and Aaghaz Mahajan and Ainesh Sanyal and Mianqi Wang and Shubhadeep Chakraborty and Sudip Chakraborty and Shailesh Chandrasekharan and Rama Govindarajan and Akshit Goyal and Jonathan J. Heckman and Yasir Iqbal and Divya Jaganathan and Saumav Kapoor and Andreas Karch and Pushkar Khandare and Sandeep Krishna and Anupam Kundu and Alok Laddha and Tongyan Lin and Swarnendu Maity and Nilmani Mathur and Shani Nadir Meynet and Alessandro Mininno and Abhijeet Mishra and Vidur Niranjan and Abhishodh Prakash and Pratik Rath and Durbar Roy and Debargha Sarkar and Srikanth Sastry and Gary Shiu and Aninda Sinha and Thomas Van Riet and Praneeth Netrapalli and Suvrat Raju},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2610.16733},
  eprint = {2610.16733},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.16733},
  note = {CC BY-SA 4.0},
}

TY  - UNPB
TI  - TheoPhysEval: Assessing AI agents at autonomous theoretical physics problem solving
AU  - A. Ramesh Chandra
AU  - Aaghaz Mahajan
AU  - Ainesh Sanyal
AU  - Mianqi Wang
AU  - Shubhadeep Chakraborty
AU  - Sudip Chakraborty
AU  - Shailesh Chandrasekharan
AU  - Rama Govindarajan
AU  - Akshit Goyal
AU  - Jonathan J. Heckman
AU  - Yasir Iqbal
AU  - Divya Jaganathan
AU  - Saumav Kapoor
AU  - Andreas Karch
AU  - Pushkar Khandare
AU  - Sandeep Krishna
AU  - Anupam Kundu
AU  - Alok Laddha
AU  - Tongyan Lin
AU  - Swarnendu Maity
AU  - Nilmani Mathur
AU  - Shani Nadir Meynet
AU  - Alessandro Mininno
AU  - Abhijeet Mishra
AU  - Vidur Niranjan
AU  - Abhishodh Prakash
AU  - Pratik Rath
AU  - Durbar Roy
AU  - Debargha Sarkar
AU  - Srikanth Sastry
AU  - Gary Shiu
AU  - Aninda Sinha
AU  - Thomas Van Riet
AU  - Praneeth Netrapalli
AU  - Suvrat Raju
PY  - 2026
DA  - 2026/10/09
PB  - arXiv
DO  - 10.48550/arxiv.2610.16733
UR  - https://arxiv.org/abs/2610.16733
AB  - We present a suite of 50 unpublished research-level theoretical physics problems with verifiable answers. Without AI assistance, a typical problem in our collection could be solved by a world expert in about a day or less but would take considerably longer for experts in adjacent specialties. We test the ability of off-the-shelf AI agents to tackle these problems autonomously, without human hints. When given access to tools available in a realistic setting, including the ability to search the literature and write scripts for numerical and symbolic work, the best-performing agent, GPT-6-Astra harnessed in Codex, obtains the final answers with a near-perfect accuracy of 98% at an average cost under USD 3 per problem. On average, agents show only small changes in accuracy when denied web search and access to background research papers, suggesting significant memorization of the literature. These results suggest that state-of-the-art agents can accurately perform concrete calculations required across theoretical physics research. We comment on the broader implications of these findings and discuss possible explanations for the current lack of agent-led breakthroughs in theoretical physics despite their outstanding problem-solving abilities.
ER  -