Terminal-Bench 2

Terminal-agent tasks in remote shell environments.

Terminal-Bench 2 evaluates agents on hard command-line tasks in task-specific workspaces.

Runtime Status

Field	Value
Benchmark id	terminal_bench_2
Tags	Terminal, Tool Use
Execution type	remote environment
Typical harness	terminus2
Typical environment	modal, daytona, or cluster provider
Current status	registered in the direct runtime

Run Example

agentcompass run \
  terminal_bench_2 \
  terminus2 \
  your-model \
  --env <env-provider> \
  --benchmark-params '{"sample_ids":["overfull-hbox"]}' \
  --model-base-url "$MODEL_BASE_URL" \
  --model-api-key "$MODEL_API_KEY"

Recipes can infer task images and use /root as the Terminal-Bench workspace root.

Outputs

Per-task details are written to results/terminal_bench_2/<model>/<run>/details/. Aggregate metrics are written to summary.md.