ResearchBenchmarksTool-use agent evaluation on τ-bench airline (test)Follow37.6Pass@1 Success RateFAMA31.77633.28834.836.312Apr 28, 2026Evaluation ResultsMethodMethodLinksPass@1 Success RatePass@2 Success RatePass@3 Success RatePass@4 Success RatePass@5 Success RateFAMABackbone=Qwen3-4B-Inst...Backbone=Qwen3-4B-Instruct-2507, Method framework=FAMA2026.0437.63228.326.726SRBackbone=Qwen3-4B-Inst...Backbone=Qwen3-4B-Instruct-2507, Method framework=SR2026.0433.223.919.616.3914ReActBackbone=Qwen3-4B-Inst...Backbone=Qwen3-4B-Instruct-2507, Method framework=ReAct2026.04322826.826.426