Alignment Evaluation on Open-ended questions
68.9Win RateSingle-Agent
Evaluation results
| Method | Links | |
|---|---|---|
| Single-AgentComparison=vs. Base, Decoding strategy=greedy2026.03 | 68.9 | |
| Single-AgentComparison=vs. Base, Decoding strategy=sampling2026.03 | 68.7 | |
| Multi-AgentComparison=vs. Base, Decoding strategy=sampling2026.03 | 63.4 | |
| Multi-AgentComparison=vs. Base, Decoding strategy=greedy2026.03 | 51.8 | |
| Multi-AgentComparison=vs. Single-Agent, Decoding strategy=sampling2026.03 | 40.4 | |
| Multi-AgentComparison=vs. Single-Agent, Decoding strategy=greedy2026.03 | 38.4 |