tau2-bench
Second version of tau-bench; rows mix domains and pass^k settings
agents and tool usecrowded
Definition
- Version
- not in the notes
- Category
- agents and tool use
- Maintainer
- Sierra
- Status
- crowded
- Home page
- https://www.benchlm.ai/benchmarks/tau2-bench
Sources
- τ²-bench results Leaderboard & Scores , October 2026 benchlm.ai https://www.benchlm.ai/benchmarks/tau2-bench
Scores
| Model | Score | Who ran it | Setting | View or population | Date | Source |
|---|---|---|---|---|---|---|
 GLM-5.2 | 99.1 % | independent | Artificial Analysis telecom run | rank 1, BenchLM aggregator, top score | 10 Oct 2026 | benchlm.ai |
 A.X K2 | 98 % | vendor-reported | Thinking mode; Telecom | A.X K2 model card | Jul 2026 | huggingface.co |
 Trinity-Large-Thinking | 94.7 % | vendor-reported | Telecom domain | Trinity-Large-Thinking model card | 1 Apr 2026 | huggingface.co |
 Motif-3 | 94.7 % | vendor-reported | temperature 1.0, top-p 0.95, max sequence 262,144; Telecom | Motif-3 model card | Aug 2026 | huggingface.co |
 LongCat-Flash (LongCat-Flash-Chat) | 73.68 % | vendor-reported | Telecom, avg@4 | LongCat-Flash-Chat model card | Sep 2025 | huggingface.co |
Where this comes from
Scores from different settings, harnesses or dates are not directly comparable; the row says which. Sorted by score, highest first, as a reading aid, not a ranking.
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "tau2-Bench", https://aurion.semai.id/benchmarks/tau2-bench/, records of October 2026.