Software Engineering Issue Resolution on SWE-Bench Multilingual
45.78Resolve RateNemotron 3 Super
Evaluation Results
| Method | Links | |
|---|---|---|
| Nemotron 3 SuperFramework/Scaffold=OpenHands2026.04 | 45.78 | |
| GPT-OSS-120BFramework/Scaffold=OpenHands2026.04 | 30.8 | |
| HybridFeedback=Hybrid (Execution-free + Execution-based)2025.12 | 0.357 | |
| Execution-based onlyFeedback=Execution-based2025.12 | 0.333 | |
| Execution-free onlyFeedback=Execution-free (SWE-RM)2025.12 | 0.33 | |
| Poor Calibrated RMFeedback=Ablated RM (Poorly Calibrated)2025.12 | 0.21 |