TauC-Bench - Airline (Gemini 3 Flash User): leaderboard

TauC-Bench (2026) extends the tau2-bench customer-service domains with tasks synthesized by the TASTE pipeline from sampled gold tool-call sequences and evolved with ambiguous or uncooperative users and decoy records. Airline domain, 50 tasks, with a Gemini 3 Flash user simulator in the tau2-bench harness. Pass^1 in percent, a task counting as solved when the final database state equals the gold state; higher is better.

Metric: pass^1 (times 100, so 0-100): mean per-task success, a task succeeding when the database state after the conversation equals the gold final state, on the 50 generated Airline tasks of TauC-Bench, run in the tau2-bench harness with default settings and a Gemini 3 Flash user simulator; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Sonnet 4.664#179
2GPT-5.249#190
3DeepSeek V3.141#511
4Gemini 2.5 Flash21#466
5Qwen 3 32B13#983

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=tauc-bench-airline-gemini-3-flash-user · How It Works · Data refreshed daily, snapshot 2026-11-22.