List your company

Multi-hop Reasoning on 2WikiMultihopQA (Acc, Len, Faith)

82.1AccuracyGeoFaith
32.59645.44858.371.152May 26, 2026

Evaluation results

MethodLinks
GeoFaithBackbone=Qwen3-4B2026.0582.10.392
GRPOBackbone=Qwen3-4B2026.0581.50.484.8
KnowRLBackbone=Qwen3-4B2026.0579.80.486.5
OriginalBackbone=Qwen3-4B2026.0579.20.690.1
TruthRLBackbone=Qwen3-4B2026.0578.10.279.5
THSBackbone=Qwen3-4B2026.0576.60.386.8
GeoFaithBackbone=Qwen3-1.7B2026.0571.20.377.5
GRPOBackbone=Qwen3-1.7B2026.0567.40.671.9
THSBackbone=Qwen3-1.7B2026.0556.50.249.3
TruthRLBackbone=Qwen3-1.7B2026.0553.60.253.9
KnowRLBackbone=Qwen3-1.7B2026.0550.50.260.3
OriginalBackbone=Qwen3-1.7B2026.0534.50.352