Multi-hop Question Answering on HotpotQA (SR)
52Success Rate (SR)Gemini-3-Flash
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini-3-FlashProtocol=Prompting2026.05 | 52 | |
| Gemini-2.5-flashProtocol=Prompting2026.05 | 47 | |
| MATPOBackbone=Qwen3-4B2026.05 | 46 | |
| Critique-GRPOBackbone=Qwen3-8B2026.05 | 46 | |
| MATPOBackbone=Qwen3-8B2026.05 | 45 | |
| ScalingInterBackbone=Qwen3-4B2026.05 | 44 | |
| ICRLBackbone=Qwen3-4B2026.05 | 44 | |
| GSPOBackbone=Qwen3-8B2026.05 | 44 | |
| ICRLBackbone=Qwen3-8B2026.05 | 44 | |
| Critique-GRPOBackbone=Qwen3-4B2026.05 | 43.5 | |
| GSPOBackbone=Qwen3-4B2026.05 | 42 | |
| ScalingInterBackbone=Qwen3-8B2026.05 | 41.5 | |
| Qwen3-30B-A3BProtocol=Prompting2026.05 | 40 | |
| GRPOBackbone=Qwen3-4B2026.05 | 38 | |
| GRPOBackbone=Qwen3-8B2026.05 | 38 | |
| Qwen3-8BProtocol=Prompting2026.05 | 32 | |
| Qwen3-4BProtocol=Prompting2026.05 | 24 |