Pokee-Isaac 28B Report - τ³-bench (text) - Airline: leaderboard
Sierra's τ³-bench text track, the airline domain (bookings, changes and cancellations): an agent follows domain policy and calls tools over many turns with a simulated user (Azure GPT-5.2 at temperature 0); seed 300, 200 maximum turns, temperature 0. pass@1 on a 0-1 scale. From Pokee AI's Pokee-Isaac 28B technical report (September 10, 2026 revision): Pokee ran every listed model itself through hosted APIs under one harness, with each peer's default temperature (unless a benchmark section fixes one) and a high thinking level where available; exhausted retries, context overflows and verifier failures count as zero. Higher is better.
Metric: pass@1 (0-1). Source: pokee.ai. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Luna | 0.72 | #117 |
| 2 | Gemini 3.5 Flash Lite | 0.7 | #300 |
| 3 | Nemotron 3 Super | 0.69 | #766 |
| 4 | Qwen 3.5 122B A10B | 0.66 | #329 |
| 5 | Claude Haiku 4.5 | 0.5 | #568 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=pokee-isaac-28b-report-tau3-bench-text-airline · How It Works · Data refreshed daily, snapshot 2026-11-12.