List your company

Multi-Hop Question Answering on MuSiQue (Exact Match (EM))

22.6Exact Match (EM)StepSearch-Instruct
1.1766.73812.317.862May 7, 2025Jul 8, 2025Sep 9, 2025Nov 11, 2025Jan 12, 2026Mar 16, 2026May 18, 2026

Evaluation results

MethodLinks
StepSearch-InstructTraining Pipeline=Pure RL, External Teacher=GPT-4o, Backbone=Qwen2.5-7B2026.0522.6
SD-Search-InstructTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.0522.2
MR-Search-BaseTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.0522.1
Thinker-InstructTraining Pipeline=Synthesize → SFT, External Teacher=Qwen2.5-72B, Backbone=Qwen2.5-7B2026.0522.1
StepSearch-BaseTraining Pipeline=Pure RL, External Teacher=GPT-4o, Backbone=Qwen2.5-7B2026.0521.6
SD-Search-BaseTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.0520.4
SkillRLEvaluation Protocol=Zero-shot2026.0520.2
SKILLGRAPHEvaluation Protocol=Zero-shot2026.0519.5
AutoRefine-BaseTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.0518.7
ZeroSearchEvaluation Protocol=Zero-shot2026.0518.4
ZEROSEARCH-instBackbone=Qwen-2.5-7B, Variant=Instruct2025.0518.4
Search-R1-baseBackbone=Qwen-2.5-7B, Variant=Base2025.0518.2
ZEROSEARCH-baseBackbone=Qwen-2.5-7B, Variant=Base2025.0518
Search-R1-instBackbone=Qwen-2.5-7B, Variant=Instruct2025.0517.4
Search-R1-BaseTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.0517.3
EvolveREvaluation Protocol=Zero-shot2026.0515.6
Search-R1Evaluation Protocol=Zero-shot2026.0514.6
Search-R1-InstructTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.0514.6
ZEROSEARCH-baseBackbone=LLaMA-3.2-3B, Variant=Base2025.0513.8
Search-R1-baseBackbone=LLaMA-3.2-3B, Variant=Base2025.0513.6
ZEROSEARCH-baseBackbone=Qwen-2.5-3B, Variant=Base2025.0513
R1-instructBackbone=LLaMA-3.2-3B, Variant=Instruct2025.0511.4
Search-R1-baseBackbone=Qwen-2.5-3B, Variant=Base2025.0511.2
ZEROSEARCH-instBackbone=LLaMA-3.2-3B, Variant=Instruct2025.0510.4
R1-baseBackbone=Qwen-2.5-7B, Variant=Base2025.059.8
ZEROSEARCH-instBackbone=Qwen-2.5-3B, Variant=Instruct2025.059.8
RAGBackbone=LLaMA-3.2-3B2025.059.6
RAGEvaluation Protocol=Zero-shot2026.059.4
RAGBackbone=Qwen-2.5-7B2025.059.4
Search-R1-instBackbone=LLaMA-3.2-3B, Variant=Instruct2025.058.8
Search-o1Evaluation Protocol=Zero-shot2026.058.6
Search-o1Backbone=Qwen-2.5-7B2025.058.6
R1-instructBackbone=Qwen-2.5-7B, Variant=Instruct2025.058.4
RAGBackbone=Qwen-2.5-3B2025.058.2
RAgentBackbone=Qwen-2.5-7B2025.057.6
Search-R1-instBackbone=Qwen-2.5-3B, Variant=Instruct2025.057.6
R1-baseBackbone=LLaMA-3.2-3B, Variant=Base2025.057
CoTEvaluation Protocol=Zero-shot2026.056.6
CoTBackbone=Qwen-2.5-7B2025.056.6
R1-baseBackbone=Qwen-2.5-3B, Variant=Base2025.056.4
Search-o1Backbone=LLaMA-3.2-3B2025.056
CoTBackbone=LLaMA-3.2-3B2025.055.8
RAgentBackbone=LLaMA-3.2-3B2025.055.6
Search-o1Backbone=Qwen-2.5-3B2025.055.2
Qwen2.5-7B-InstructEvaluation Protocol=Zero-shot2026.054.8
Direct AnswerBackbone=Qwen-2.5-7B2025.054.8
Direct AnswerBackbone=Qwen-2.5-3B2025.054.4
R1-instructBackbone=Qwen-2.5-3B, Variant=Instruct2025.054.4
CoTBackbone=Qwen-2.5-3B2025.053.6
RAgentBackbone=Qwen-2.5-3B2025.052.6
Direct AnswerBackbone=LLaMA-3.2-3B2025.052