Multi-hop Question Answering on 2WikiMultiHopQA (EM, Tok)
48.3EMCARL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CARLModel Scale=7B-Inst.2026.05 | 48.3 | 1,139 | |
| CARLModel Scale=3B-Inst.2026.05 | 44.7 | 1,394 | |
| IGRPOBackbone=Qwen2.5-7B-Instruct, Type=Tree-based2026.07 | 43.8 | — | |
| GiGPOBackbone=Qwen2.5-7B-Instruct, Type=Chain-based2026.07 | 43.6 | — | |
| IGPOBackbone=Qwen2.5-7B-Instruct, Type=Chain-based2026.07 | 42.5 | — | |
| Search-R1 PPOModel Scale=7B-Inst.2026.05 | 41.7 | 1,775 | |
| IGRPOBackbone=Qwen2.5-3B-Instruct, Type=Tree-based2026.07 | 40.2 | — | |
| Search-R1Backbone=Qwen2.5-7B-Instruct, Type=Chain-based2026.07 | 40.1 | — | |
| GiGPOBackbone=Qwen2.5-3B-Instruct, Type=Chain-based2026.07 | 37 | — | |
| Search-R1 GRPOModel Scale=7B-Inst.2026.05 | 36.8 | 1,827 | |
| StepSearchBackbone=Qwen2.5-7B-Instruct, Type=Chain-based2026.07 | 36.6 | — | |
| Tree-GRPOBackbone=Qwen2.5-7B-Instruct, Type=Tree-based2026.07 | 35.6 | — | |
| ZeroSearchBackbone=Qwen2.5-7B-Instruct, Type=Chain-based2026.07 | 35.2 | — | |
| IGPOBackbone=Qwen2.5-3B-Instruct, Type=Chain-based2026.07 | 34.9 | — | |
| Search-R1 GRPOModel Scale=3B-Inst.2026.05 | 34.3 | 1,960 | |
| SFT (rej. sampled)Model Scale=7B-Inst.2026.05 | 32.6 | 1,683 | |
| Search-R1 PPOModel Scale=3B-Inst.2026.05 | 32.1 | 1,861 | |
| StepSearchBackbone=Qwen2.5-3B-Instruct, Type=Chain-based2026.07 | 32 | — | |
| Search-R1Backbone=Qwen2.5-3B-Instruct, Type=Chain-based2026.07 | 31.9 | — | |
| Tree-GRPOBackbone=Qwen2.5-3B-Instruct, Type=Tree-based2026.07 | 31.7 | — | |
| ZeroSearchBackbone=Qwen2.5-3B-Instruct, Type=Chain-based2026.07 | 30 | — | |
| R1-InstructBackbone=Qwen2.5-3B-Instruct, Type=RL Training2026.07 | 29.2 | — | |
| R1-InstructBackbone=Qwen2.5-7B-Instruct, Type=RL Training2026.07 | 27.5 | — | |
| CoT+Tools (selective)Model Scale=7B-Inst.2026.05 | 27.1 | 1,641 | |
| CoT+Tools (always)Model Scale=7B-Inst.2026.05 | 26.6 | 1,938 | |
| SFT (rej. sampled)Model Scale=3B-Inst.2026.05 | 25.1 | 1,791 | |
| CoT+Tools (selective)Model Scale=3B-Inst.2026.05 | 21.2 | 1,732 | |
| CoT+Tools (always)Model Scale=3B-Inst.2026.05 | 19.4 | 2,028 |