Multi-hop Question Answering on MuSiQue (SR)
31Success Rate (SR)Gemini-3-Flash
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini-3-FlashProtocol=Prompting2026.05 | 31 | |
| Gemini-2.5-flashProtocol=Prompting2026.05 | 26 | |
| MATPOBackbone=Qwen3-8B2026.05 | 17 | |
| Qwen3-30B-A3BProtocol=Prompting2026.05 | 16 | |
| MATPOBackbone=Qwen3-4B2026.05 | 16 | |
| ICRLBackbone=Qwen3-4B2026.05 | 16 | |
| ICRLBackbone=Qwen3-8B2026.05 | 16 | |
| ScalingInterBackbone=Qwen3-4B2026.05 | 15 | |
| Critique-GRPOBackbone=Qwen3-4B2026.05 | 15 | |
| GSPOBackbone=Qwen3-8B2026.05 | 15 | |
| GRPOBackbone=Qwen3-4B2026.05 | 14 | |
| GRPOBackbone=Qwen3-8B2026.05 | 14 | |
| ScalingInterBackbone=Qwen3-8B2026.05 | 14 | |
| Critique-GRPOBackbone=Qwen3-8B2026.05 | 14 | |
| GSPOBackbone=Qwen3-4B2026.05 | 10 | |
| Qwen3-4BProtocol=Prompting2026.05 | 6 | |
| Qwen3-8BProtocol=Prompting2026.05 | 6 |