List your company
ResearchFormerly WizWand

Multi-turn Decision Making on tau2-Bench held-out (test)

41.9Average ScoreCURATEEVO
23.90828.57933.2537.921Jul 7, 2026
Updated 4mo ago

Evaluation results

MethodLinks
CURATEEVOModel=QWEN3-4B, Data Setting=Labeled Data2026.0741.9
EnvScalerModel=QWEN3-8B, Data Setting=Labeled Data2026.0737.9
CURATEEVOModel=QWEN3-4B, Data Setting=Wild Data2026.0737.2
EnvScalerModel=QWEN3-8B, Data Setting=Wild Data2026.0735.5
AWMModel=QWEN3-8B, Data Setting=Labeled Data2026.0733.5
MUA-RLModel=QWEN3-8B, Data Setting=Labeled Data2026.0732.8
GRPO w/o. Data CurationModel=QWEN3-8B, Data Setting=Labeled Data2026.0731.8
AWMModel=QWEN3-8B, Data Setting=Wild Data2026.0731.8
RODSModel=QWEN3-4B, Data Setting=Labeled Data2026.0731.6
MUA-RLModel=QWEN3-8B, Data Setting=Wild Data2026.0730.6
FunReason-MTModel=QWEN3-4B, Data Setting=Labeled Data2026.0730.2
GRPO w/o. Data CurationModel=QWEN3-8B, Data Setting=Wild Data2026.0729.6
RODSModel=QWEN3-4B, Data Setting=Wild Data2026.0729.6
FunReason-MTModel=QWEN3-4B, Data Setting=Wild Data2026.0727.8
GRPO w/o. Data CurationModel=QWEN3-4B, Data Setting=Labeled Data2026.0726.8
GRPO w/o. Data CurationModel=QWEN3-4B, Data Setting=Wild Data2026.0724.6