MongoDB - Build AI That Scales
Mini toolsTau2-Bench Telecom
τ²-Bench Telecom
Multi-turn customer-service agent benchmark in a telecom domain: the model must take real tool actions while a simulated customer pushes back on incomplete or wrong answers.
Agentic
Text
Accuracy
Max 100.0%
Released Jun 2025
12
Results
12
Models scored
99.3%
Top: Gemini 3.1 Pro
93.5%
Median
Best results
Top primary scores; one row per model.
Frontier over time
Each dot is one model result; the line traces the running best score.
All results
Showing all configurations including non-primary alternates. · Show only primary
| # | Model | Score | Conditions | Eval date | Source | Flags |
|---|---|---|---|---|---|---|
| 1 | Gemini 3.1 Pro | 99.3% | CoT | 19 Feb 2026 | Self-reported | Primary |
| 2 | GPT 5.4 | 98.9% | — | 05 Mar 2026 | Self-reported | Primary |
| 3 | Claude Opus 4.5 | 98.2% | — | 24 Oct 2025 | Self-reported | Primary |
| 4 | GPT 5.5 | 98.0% | CoT | 23 Apr 2026 | Self-reported | Primary |
| 5 | Claude Sonnet 4.6 | 97.9% | — | 17 Feb 2026 | Self-reported | Primary |
| 6 | GPT 5 (Thinking) | 96.7% | — | 07 Aug 2025 | Self-reported | Primary |
| 7 | Gemini 3 Flash (Thinking) | 90.2% | — | 17 Dec 2025 | Self-reported | Primary |
| 8 | Claude Haiku 4.5 | 83.0% | — | 15 Oct 2025 | Self-reported | Primary |
| 9 | Gemini 2.5 Flash (Thinking) | 79.5% | — | 17 Dec 2025 | Self-reported | Primary |
| 10 | Qwen 3.5 27B | 79.0% | — | 24 Feb 2026 | Third-party | Primary Verified |
| 11 | Gemini 2.5 Pro (Thinking) | 77.8% | — | 17 Dec 2025 | Self-reported | Primary |
| 12 | Claude Sonnet 3.5 | 62.6% | 0-shot · standard | 22 Oct 2024 | Self-reported | |
| 13 | GPT 5 | 38.6% | — | 07 Aug 2025 | Self-reported | Primary |
| 14 | Claude Haiku 3 | 18.2% | 0-shot · standard | 22 Oct 2024 | Self-reported |
