Command-line Task Execution on Terminal-Bench 2 (held-out)
76Pass RateRHO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RHOHarness Architecture=Skills+Tools2026.06 | 76 | 0.05 | |
| Dynamic CheatsheetHarness Architecture=Skills2026.06 | 73 | 0.02 | |
| ReasoningBankHarness Architecture=Memory2026.06 | 73 | 0.02 | |
| Sleep-time ComputeHarness Architecture=Memory2026.06 | 73 | 0.02 | |
| Vanilla CodexHarness Architecture=None2026.06 | 71 | — |