Can AI Agents Do Real Science Research? — Steven Dillmann, Stanford

Frontier Data Summit 2026: Terminal-Bench-Science tests whether AI agents can do real scientific research work, with scientists leading the task review. Terminal-Bench-Science evaluates AI agents on scientific research workflows across life, physical, Earth, mathematical, and engineering sciences. This talk covers the scientist-led task review process, examples from the 0.1 release, early model results, and how researchers can contribute tasks. The project received support through Snorkel AI’s Open Benchmarks Grants. IN THIS VIDEO • The five scientific domains the benchmark covers • Task criteria and the scientist-led review pipeline • Examples and results from the 0.1 release • Leaderboard results across models and domains • How researchers can contribute tasks SPEAKER Steven Dillmann, PhD Student, Stanford; Project Lead, Terminal-Bench-Science CHAPTERS 00:00 Introducing Terminal-Bench-Science 02:06 Five scientific domains and the project’s goal 02:46 Task criteria and the review pipeline 04:11 First release results and task examples 05:36 Math, engineering, and benchmark performance 06:21 Leaderboard results across models and domains 07:39 Roadmap and invitation to contribute RESOURCES Benchmark, tasks, and leaderboard: https://www.terminal-bench-science.ai/ Snorkel AI Open Benchmarks Grants: https://benchmarks.snorkel.ai/