Four-Factor Integrity Breakdown & Primary-Source Evidence
Access Model: Public Plaintext
"Terminal-Bench 4.1 Agentic terminal use, 66 tasks By Laude Institute"
"The official published dataset and leaderboard are hosted on the Harbor Hub. Usage e.g. harbor run -d terminal-bench/terminal-bench@4.1.0 ."
Originality
Weight 25%
10/10
Reuse Profile: All Original (0% reuse)
"We improved verifiers for reliability and robustness to reward hacking. Many reported exploits remained theoretical and were not seen on the leaderboard. We re-graded TB 4.0 with these fixes, and all scores remain within previously reported 95% CIs."
Freshness
Weight 20%
7/10
Initial Release: 2025-05Latest Refresh: 2026-11Rolling Updates: No
"Terminal-Bench 4.1 implements verifier fixes, reproducibility improvements, hardening against reward hacking. Also implements a new CPU-only dataset containing 63 tasks."
Grading
Weight 15%
3.3/10
Integrity: Self-Reported / Live NetworkSandbox: live internet
"allow_internet = true # Terminal-Bench is open internet"
Benchmark maintainers can embed this live SVG badge in GitHub READMEs or documentation:
[](https://benchaudit.org/benchmarks/terminal-bench-4-1/)