List your company

Tool-use agent evaluation on τ-bench retail

44.173Pass@1 Success RateFAMA
26.0010830.7187935.436540.15421Apr 28, 2026

Evaluation results

MethodLinks
FAMABackbone=Qwen2.5-72B-Instruct2026.0444.17334.2630.2628.1726.95
ReActBackbone=Qwen2.5-72B-Instruct2026.0443.4732.2626.6923.1320.86
SRBackbone=Qwen2.5-72B-Instruct2026.0442.930.72521.218.2
FAMABackbone=Qwen3-32B2026.0440.526.919.915.312.2
ReActBackbone=Qwen3-32B2026.04382518.815.310
SRBackbone=Qwen3-32B2026.0426.715.611.39.27.8