Wikibench

TAU-bench Airline

reasoning

Part of τ-bench (TAU-bench), a benchmark for Tool-Agent-User interaction in real-world domains. The airline domain evaluates language agents' ability to interact with users through dynamic conversations while following domain-specific rules and using API tools. Agents must handle airline-related tasks and policies reliably.

Leaderboard

Showing 20 of 23 results

  1. Claude Sonnet 4.5
    70.0%
    i
  2. MiniMax M1 80K
    62.0%
    i
  3. GLM-4.5-Air
    60.8%
    i
  4. GLM-4.5
    60.4%
    i
  5. Claude Sonnet 4
    60.0%
    i
  6. MiniMax M1 40K
    60.0%
    i
  7. Qwen3-Coder 480B A35B Instruct
    60.0%
    i
  8. Claude Opus 4
    59.6%
    i
  9. Claude 3.7 Sonnet
    58.4%
    i
  10. Claude Opus 4.1
    56.0%
    i
  11. GPT-4.5
    50.0%
    i
  12. o1
    50.0%
    i
  13. GPT-4.1
    49.4%
    i
  14. o4-mini
    49.2%
    i
  15. Qwen3-Next-80B-A3B-Thinking
    49.0%
    i
  16. Claude 3.5 Sonnet
    46.0%
    i
  17. Qwen3-235B-A22B-Thinking-2507
    46.0%
    i
  18. Qwen3-Next-80B-A3B-Instruct
    44.0%
    i
  19. GPT-4o
    42.8%
    i
  20. GPT-4.1 mini
    36.0%
    i