| Benchmark | Model | Rank | Results |
|---|---|---|---|
| agents-on-bench | Qwen3-30B-A3B-Thinking-2507 + PivotRL | – | Accuracy: 63.81 |
| code-generation-on-swe-bench-verified | Qwen3-30B-A3B-Thinking-2507 + PivotRL (OpenHands) | – | Accuracy: 32.67 |
| deep-research-agents-on-browsecomp | Qwen3-30B-A3B-Thinking-2507 + PivotRL | – | Accuracy: 11.3 |