List your company

Tool-use Task Completion on Tau-Bench Airline v2 (test)

70Pass RateClaude Sonnet 4.5
11.7626.884257.12Apr 3, 2026

Evaluation results

MethodLinks
Claude Sonnet 4.52026.0470—
Qwen3-30B-A3B + IRCSize=30.5B2026.0469.511.5
Qwen3-30B-A3B + MT-GRPOSize=30.5B2026.046810
Qwen3.5-4B + IRCSize=4B2026.0466.72.9
Qwen3.5-4B + MT-GRPOSize=4B2026.0464.60.8
Qwen3.5-4B (base)Size=4B2026.0463.8—
Qwen3-30B-A3B (base)Size=30.5B2026.0458—
GPT-4.12026.0449.4—
GPT-4o2026.0442.8—
Claude 3.5 Haiku2026.0422.8—
GPT-4.1 nano2026.0414—