Reasoning on GPQA Protocol A (test)
87.3AccuracyOpenHands CodeActAgent + GBT-SE
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| OpenHands CodeActAgent + GBT-SEBackbone=gpt-4o, Method=GBT-SE2026.01 | 87.3 | 73 | 0.2 | 0 | 15 | 58 | |
| OpenHands CodeActAgent + GBT-BasicBackbone=gpt-4o, Method=GBT-Basic2026.01 | 78.8 | 71.9 | 0.2 | 0 | 16 | 62 | |
| OpenHands CodeActAgent + Global guardrail onlyBackbone=gpt-4o, Configuration=Global guardrail only2026.01 | 59.2 | — | 0.4 | 20 | 21 | 82 | |
| OpenHands CodeActAgentBackbone=gpt-4o, Mode=native2026.01 | 58.7 | — | 1.6 | 40 | 22 | 86 | |
| Zero-shot promptingBackbone=gpt-4o, Mode=zero-shot2026.01 | 53.6 | — | — | — | — | — |