List your company
ResearchFormerly WizWand

Knowledge-Intensive Reasoning on 2WikiMultiHopQA

81.5AccuracyAPPO
7.55626.75345.9565.147Dec 15, 2025Jan 13, 2026Feb 12, 2026Mar 13, 2026Apr 12, 2026May 11, 2026Jun 10, 2026
Updated 4mo ago

Evaluation results

MethodLinks
APPOBackbone=Qwen2.5-7B-Instruct2026.0681.5
APPOBackbone=Llama3.1-8B-Instruct2026.0679.8
GRPOBackbone=Qwen2.5-7B-Instruct2026.0676.1
ARPOBackbone=Qwen2.5-7B-Instruct2026.0676.1
CISPOBackbone=Llama3.1-8B-Instruct2026.0675.6
ARPOBackbone=Llama3.1-8B-Instruct2026.0675.5
GIGPOBackbone=Llama3.1-8B-Instruct2026.0674.6
GPPOBackbone=Qwen2.5-7B-Instruct2026.0674.2
GIGPOBackbone=Qwen2.5-7B-Instruct2026.0673.5
GPPOBackbone=Llama3.1-8B-Instruct2026.0672.8
CISPOBackbone=Qwen2.5-7B-Instruct2026.0672.1
GRPOBackbone=Llama3.1-8B-Instruct2026.0671.8
Reinforce++Backbone=Llama3.1-8B-Instruct2026.0671.6
DAPOBackbone=Llama3.1-8B-Instruct2026.0670.3
Reinforce++Backbone=Qwen2.5-7B-Instruct2026.0668.9
DAPOBackbone=Qwen2.5-7B-Instruct2026.0668.4
AutoTool (Qwen3-8B)Size=8B, Backbone=Qwen32025.1248.8
TIR PromptingBackbone=Llama3.1-8B-Instruct2026.0647.5
Search-R1Size=7B2025.1244.5
AutoTool (Qwen2.5-VL-7B)Size=7B, Backbone=Qwen2.5-VL2025.1236.5
GPT4oSize=-2025.1229.8
Tool-StarBackbone=Qwen2.5-7B, Training Paradigm=SFT–RL TIR, SFT Trajectory Count=54K2026.0129.5
AutoTrajBackbone=Qwen2.5-7B, Training Paradigm=SFT–RL TIR, SFT Trajectory Count=13K2026.0129.5
Vanilla SFT-RL TIRBackbone=Qwen2.5-7B, Training Paradigm=SFT–RL TIR2026.0128.5
ReSearchBackbone=Qwen2.5-7B, Training Paradigm=RL-only2026.0126.5
AutoTIRBackbone=Qwen2.5-7B, Training Paradigm=RL-only2026.0125
Zero-shotBackbone=Llama3.1-8B-Instruct2026.0624.6
QwQ-32BSize=32B2025.1222.3
Tool-Star-SFTBackbone=Qwen2.5-7B, Training Paradigm=SFT-only, SFT Trajectory Count=54K2026.0121.5
Qwen2.5-7B-InstructBackbone=Qwen2.5-7B, Training Paradigm=Instruct2026.0120.5
ReToolSize=32B2025.1219.7
TIR PromptingBackbone=Qwen2.5-7B-Instruct2026.0618.3
R1-SearcherBackbone=Qwen2.5-7B, Training Paradigm=RL-only2026.0117
Qwen2.5-Math-72B-InstructSize=72B2025.1216.7
DeepSeek-R1-Distill-Qwen-14BSize=14B2025.1214.1
ToRLBackbone=Qwen2.5-7B, Training Paradigm=RL-only2026.0114
Zero-shotBackbone=Qwen2.5-7B-Instruct2026.0612.6
Qwen2.5-VL-72B-InstructSize=72B2025.1210.4