A benchmark to measure and evolve with the frontier of agent work.

89 high-quality tasks across software engineering, machine learning, security, data science, and more.

The original Terminal-Bench benchmark. 80 tasks testing agents' abilities to complete tasks using a terminal.

A domain-specific benchmark for scientific computing in terminal environments. Currently in development.

Single-task Terminal-Bench challenges spanning inference engine code golf, Rust compiler speedup, and WASM rendering.

A lightweight, diverse, difficult, and high-quality benchmark for agentic evaluation. 82 tasks distilled from 6,627 candidates across 54 benchmarks integrated as Harbor adapters.