Software Engineering on SWE-Bench Pro (held-out)
78Pass RateRHO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RHOHarness Architecture=Skills+Tools2026.06 | 78 | 0.19 | |
| Sleep-time ComputeHarness Architecture=Memory2026.06 | 64 | 0.05 | |
| Dynamic CheatsheetHarness Architecture=Skills2026.06 | 62 | 0.03 | |
| ReasoningBankHarness Architecture=Memory2026.06 | 61 | 0.02 | |
| Vanilla CodexHarness Architecture=None2026.06 | 59 | — |