tau2-bench
2 models tested · Updated 2026-03-02 · Verified sources only
Qwen3.5 4B leads at 79.9%
1
Qwen · HuggingFace/Qwen · 2026-03-02
Edges out the 9B on tau2-bench agentic tool-use.
79.9%
2
Alibaba · HuggingFace/Qwen · 2026-03-02
Strong agentic tool-use; near-top among open sub-10B models.
79.1%