SEATauBench (L2 Interaction) - Indonesian Airline: leaderboard
The 50 tau2-bench airline customer-service tasks localized for Southeast Asian languages (SEATauBench, Nguyen et al., 2026): the simulated user and the agent converse in Indonesian, while tools, policy and database stay in English. pass@1 in percent, the mean success over three trials of reaching the expected final state, with a Qwen3-235B-A22B-Instruct-2507 user simulator; higher is better.
Metric: pass@1 (%; mean success over three trials per task of reaching the expected final state; the 50 tau2-bench airline tasks with the simulated user and the agent converse in Indonesian; tools, policy and database stay in English; Qwen3-235B-A22B-Instruct-2507 user simulator, temperature 0). Source: arxiv.org. Saturation forecast: Around February 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2.5 | 74 | #252 |
| 2 | GPT-5 Mini | 62.7 | #349 |
| 3 | Qwen 3 235B A22B 2507 Instruct | 49.3 | #619 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=seataubench-l2-interaction-indonesian-airline · How It Works · Data refreshed daily, snapshot 2026-11-12.