Software Engineering on SWE-bench Verified 500 issues (Protocol A)
73.6Success Rate (SR)OpenHands CodeAct + GBT-SE
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| OpenHands CodeAct + GBT-SEBackbone=gpt-4o, Protocol=A2026.01 | 73.6 | 86 | 0.2 | 0 | 126 | 490 | |
| OpenHands CodeAct + GBT-BasicBackbone=gpt-4o, Protocol=A2026.01 | 50.2 | 84.4 | 0.4 | 0.2 | 148 | 570 | |
| OpenHands CodeAct + Global guardrail onlyBackbone=gpt-4o, Protocol=A2026.01 | 38.8 | — | 0.8 | 0.2 | 196 | 770 | |
| OpenHands CodeAct (gpt-4o, native)Backbone=gpt-4o, Protocol=A2026.01 | 34.6 | — | 2.8 | 1.2 | 208 | 820 |