Benchmark

Terminus

Terminal-Bench is a benchmark for testing AI agents in real terminal environments, evaluating how well agents can handle real-world, end-to-end tasks autonomously. The benchmark includes tasks spanning coding, system administration, security, data science, model training, file operations, version control, and web development. Terminus is the neutral test-bed agent designed to work with Terminal-Bench, operating purely through tmux sessions without dedicated tools.