Software Engineering on SWE-bench Verified (decidable subset)
286n_mixRandom rollout selection
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Random rollout selectionModel=gpt-oss-20B, Anchor=<|start|>, N (rollouts)=162026.06 | 286 | 52 | — | — | |
| Routing Agreement Decoding (RAD)Model=gpt-oss-20B, Anchor=<|start|>, N (rollouts)=162026.06 | 286 | — | 57 | 5 | |
| Random rollout selectionModel=gpt-oss-120B, Anchor=<|start|>, N (rollouts)=162026.06 | 247 | 58.7 | — | — | |
| Routing Agreement Decoding (RAD)Model=gpt-oss-120B, Anchor=<|start|>, N (rollouts)=162026.06 | 247 | — | 63.6 | 4.8 | |
| Random rollout selectionModel=Qwen3.6-35B-A3B, Anchor=</think>, N (rollouts)=162026.06 | 233 | 55.8 | — | — | |
| Routing Agreement Decoding (RAD)Model=Qwen3.6-35B-A3B, Anchor=</think>, N (rollouts)=162026.06 | 233 | — | 60.5 | 4.7 |