Skip to content

τ³-Bench Tool-Agent-User Evaluation

τ³-bench is the current evolution of Sierra's tool-agent-user framework, adding corrected task releases and newer knowledge and voice evaluation modes alongside airline, retail, and telecom.

23 models

Higher is better.

  1. 1Mercury 2.5Inception96.0
  2. 2Mistral Medium 3.5 128BMistral91.4
  3. 3MiMo-V2.5-ProXiaomi72.9
  4. 4Nemotron 3 UltraNVIDIA70.9
  5. 5Qwen3.6 PlusAlibaba70.7
  6. 6GLM-5.1Z.AI70.6
  7. 7Claude Opus 4.5Anthropic70.2
  8. 8Qwen3.5 397BAlibaba68.4
  9. 9MiniMax M2.7MiniMax67.6
  10. 10Qwen3.6-35B-A3BAlibaba67.2
  11. 11Pokee-Isaac 28BPokee AI66.2
  12. 12Kimi K2.5Moonshot AI65.7
  13. 13GLM-5Z.AI65.6
  14. 14Granite 4.2 8BIBM58.1
  15. 15Ling 3.1 FlashInclusionAI47.8
  16. 16Atria Dawn PreviewShanghai Artificial Intelligence Laboratory41.2
  17. 17Beam 1.0 501B-A23BReflection AI36.1
  18. 18Mercury 3Inception30.2
  19. 19Nemotron 3.5 Super VL 120B A12BNVIDIA18.9
  20. 20Nemotron 3.5 Bolt 74B-A9BNVIDIA17.0
  21. 21Nemotron 3.5 Lightning 30B A3B NVFP4NVIDIA9.5
  22. 22LFM2.5-24B-A2BLiquid AI8.0
  23. 23LFM2.5-2.6BLiquid AI5.7
Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Published domain or average success results
Tasks
Corrected customer-service tasks plus knowledge and voice evaluation modes
Authors
Sierra Research (2026)

Scores are as reported by model developers or evaluators and may use different settings (for example reasoning effort or tools). Compare with care.