Released Terminal-Bench-Science 0.1, a benchmark to evaluate AI agents on scientific research workflows.