GoKawiilNew Terminal-Bench-Science benchmark shows top AI agent solving just 30% of research tasks
Stanford researchers, working with the Terminal-Bench team and scientists across disciplines, launched Terminal-Bench-Science, a benchmark testing AI agents on real scientific research workflows. The initial version includes 70 tasks spanning life, physical, Earth, mathematical, and engineering sciences, with Claude Opus 5 running Claude Code topping the leaderboard at a 30% resolution rate.
GoKawiil's interpretation of the reporting above, not reported fact.
The benchmark shifts the standard for measuring AI's scientific usefulness away from model developers and toward practicing scientists who define what tasks actually matter in research. Low resolution rates across all tested models, even the best performer solving less than a third of tasks, show current AI agents remain far from reliably automating complex scientific workflows, tempering expectations about near-term AI-driven research acceleration.
- Terminal-Bench-Science tests AI agents on 70 real scientific workflow tasks across multiple disciplines.
- Claude Opus 5 with Claude Code leads the leaderboard but resolves only 30% of tasks.
- The benchmark aims to guide AI development toward assisting, not replacing, human judgment in research.
O'Reilly "Python for Data Analysis" by Wes McKinney — If a benchmark of real research workflows has you curious about the command-line and data-wrangling skills scientists actually use, this is the classic hands-on guide to pandas, NumPy and Jupyter. It's a practical way to build the same terminal-driven analysis pipelines that AI agents are now being tested on.
See O'Reilly "Python for Data Analysis" by Wes McKinney on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.Source: terminal-bench-science.ai, 2026-08-28
Published there as: “Terminal-Bench-Science: Evaluating AI agents on scientific research workflows”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.