Software Engineering

Terminal-Bench 2.1

Terminal-Bench 2.1 tests whether an agent can finish a real task inside a Linux container it drives through a shell, passing only when a test suite run against the state it leaves behind succeeds entirely.

89items
7subjects
Apache-2.0license
software_engineeringdomain
agents_and_tool_usedomain
textmodality
item-level responses released
Saturation status: No

Response matrix

Rasch analysis p = σ(θ − z + c)

5,785 responses, 80/20 split over cells · 7 subjects · 89 items · 10 conditions

AUC train
0.900
AUC test
0.883

Loading session strips…

Correct (1)Incorrect (0)Unobserved

Scale: 1 = correct · 0 = incorrect