AA TAU-2 Bench — leaderboard

Artificial Analysis independent evaluation of TAU-2 Bench: dual-control conversational AI in telecom support scenarios (114 tasks).

Metric: Accuracy (%). Source: artificialanalysis.ai. Status: saturated. 439 models tracked.

Top models

#ModelScore
1GLM-5.2 (Max)99.12
2JT-35B-Flash99.12
3GLM-5-Turbo98.54
4Step 3.7 Flash98.54
5GLM-5 (Thinking)98.25
6Qwen 3.6 Plus97.66
7GLM-5.1 (Thinking)97.66
8Grok 4.3 (High)97.66
9GLM-5 (Non-reasoning)97.37
10GLM-5.1 (Non-reasoning)97.08
11Grok 4.20 0309 (Reasoning)96.49
12DeepSeek V4 Pro (Reasoning, Max Effort)96.2
13Kimi K2.695.91
14Qwen 3.6 Max Preview95.91
15Kimi K2.5 (Thinking)95.91

Interactive version: theaggregate.ai/benchmark?slug=aa-tau-2-bench · How the rankings work · Data refreshed daily, snapshot 2026-07-22.