BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases9

τ²-Bench Telecom (AA)

Artificial Analysis' τ²-Bench telecom run. Shown for reference.

As of 25 Oct 2026, GLM 5.2 leads τ²-Bench Telecom (AA) on BenchLeader with 99.1%, ahead of JT-35B-Flash at 99.1%, across 413 model configurations with a published result.

Published by
Artificial Analysis
Category
Agents & tools
Index weight
Reference only
Models
413
Data as of
25 Oct 2026
Newest result
9 Jul 2026

The publisher has posted no new results on this board since 9 Jul 2026. BenchLeader re-reads it every day, so newer models appear here as soon as they are published.

Source: Artificial Analysis (artificialanalysis.ai). Data taken from the public leaderboard.

What the test looks like

τ²-bench simulates customer-service conversations in airline, retail and telecom domains, where the model must follow policy, use tools and coordinate with a simulated user who also takes actions.

How it is scored

Percent of conversations ending in the correct database state.

What to keep in mind

Artificial Analysis' run of the telecom domain of τ²-bench, where the model acts as a support agent using tools while a simulated customer also acts. One domain only, so it is narrower than the full τ²-bench board. Shown for reference. See also the τ²-bench board.

Top 15
  1. 1GLM 5.2 (max)99.1%
  2. 2JT-35B-Flash99.1%
  3. 3GLM 4.7 Flash (thinking)98.8%
  4. 4Claude Fable 5 (max)98.5%
  5. 5GLM 5 Turbo98.5%
  6. 6GLM 5V Turbo (thinking)98.5%
  7. 7Step 3.7 Flash98.5%
  8. 8GLM-5 (thinking)98.3%
  9. 9Qwen3.6 Plus97.7%
  10. 10Grok 4.3 (high)97.7%
  11. 11GLM 5.1 (thinking)97.7%
  12. 12GLM-5 (none)97.4%
  13. 13GLM 5.1 (none)97.1%
  14. 14DeepSeek V4 Pro 0424 (max)96.2%
  15. 15Qwen3.6 Max (max)95.9%
413 of 413
#
1GLM 5.2max99.1%62.32026-06-16
2JT-35B-Flash99.1%50.32026-05-14
3GLM 4.7 Flashthinking98.8%47.72026-01-19
4Claude Fable 5max98.5%69.32026-06-09
5GLM 5 Turbo98.5%55.52026-03-15
6GLM 5V Turbothinking98.5%53.92026-04-01
7Step 3.7 Flash98.5%51.52026-05-29
8GLM-5thinking98.3%58.92026-02-11
9Qwen3.6 Plus97.7%58.82026-04-02
10Grok 4.3high97.7%57.02026-04-30
11GLM 5.1thinking97.7%56.62026-04-07
12GLM-5no reasoning97.4%52.32026-02-11
13GLM 5.1no reasoning97.1%50.72026-04-07
14DeepSeek V4 Pro 0424max96.2%59.62026-04-24
15Qwen3.6 Maxmax95.9%62.32026-04-20
16Kimi K2.6thinking95.9%59.42026-04-20
17Kimi K2.5thinking95.9%57.62026-01-27
18GLM 4.7thinking95.9%52.12025-12-22
19Gemini 3.5 Flashmedium95.6%62.82026-05-19
20Gemini 3.1 Pro95.6%60.52026-02-19
21DeepSeek V4 Flash 0420high95.6%55.42026-04-24
22Qwen3.5 397B A17Bthinking95.6%55.12026-02-16
23Gemini 3.5 Flashhigh95.3%63.12026-05-19
24MiniMax M2.595.3%53.62026-02-12
25Qwen3.6 35B A3Bthinking95.3%52.22026-04-16
26Mimo v2 Pro95.0%58.62026-03-18
27DeepSeek V4 Flash 0420max95.0%57.02026-04-24
28MiMo-V2-Flashthinking95.0%52.42025-12-16
29Qwen3.7max94.7%60.62026-05-19
30Claude Opus 4.8max94.4%63.32026-05-28
31Step 3.5 Flash 020294.4%50.72026-02-02
32DeepSeek V4 Flash 0420no reasoning94.4%49.02026-04-24
33DeepSeek V4 Pro 0424high94.2%57.82026-04-24
34MiMo-V2.5-Prothinking94.2%57.02026-04-22
35Qwen3.6 27Bthinking94.2%53.92026-04-22
36Mistral Medium 3.594.2%49.42026-04-29
37GLM 4.7no reasoning94.2%48.92025-12-22
38GPT-5.5xhigh93.9%66.12026-04-23
39Qwen3.5-27Bthinking93.9%54.72026-02-24
40Kimi K2.6no reasoning93.9%53.32026-04-20
41Qwen3.5-122B-A10Bthinking93.6%52.62026-02-24
42Qwen3.6 27Bno reasoning93.6%47.92026-04-22
43Grok 4.1 Fastthinking93.3%54.72025-11-19
44MiMo-V2-Flash93.3%51.62025-12-16
45Tri-21B-Thinkthinking93.3%43.42026-02-10
46GPT-5.5high93.0%65.32026-04-23
47Grok 4.20thinking93.0%58.62026-04-07
48Qwen3.7 Plus93.0%58.52026-06-01
49Kimi K2 Thinkingthinking93.0%52.42025-11-06
50JT-MINI93.0%42.92026-04-15
51Nova 2.0 Pro Previewmedium92.7%50.52025-11-27
52Ring-2.6-1T92.4%49.32026-05-08
53GPT-5.2-Codexxhigh92.1%59.42025-12-11
54Claude Opus 4.6max92.1%57.42026-02-05
55Qwen3.5-4Bthinking92.1%46.62026-03-02
56GPT-5.5medium91.8%62.92026-04-23
57GLM 4.7 Flashno reasoning91.8%41.02026-01-19
58Muse Spark91.5%64.32026-04-08
59Grok 4.3medium91.2%58.72026-04-30
60DeepSeek V4 Pro 0424no reasoning91.2%50.82026-04-24

More agents & tools benchmarks · All LLM benchmarks

Cite as: BenchLeader, “τ²-Bench Telecom (AA) leaderboard”, data as of 25 Oct 2026, https://www.benchleader.com/benchmarks/aa_tau2.

τ²-Bench Telecom (AA): questions

What does τ²-Bench Telecom (AA) measure?
τ²-bench simulates customer-service conversations in airline, retail and telecom domains, where the model must follow policy, use tools and coordinate with a simulated user who also takes actions. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads τ²-Bench Telecom (AA)?
GLM 5.2 leads τ²-Bench Telecom (AA) with 99.1% as of 25 Oct 2026, ahead of JT-35B-Flash at 99.1%.
How many models have τ²-Bench Telecom (AA) results?
413 model configurations have a τ²-Bench Telecom (AA) result on BenchLeader, all taken from Artificial Analysis.
Who runs τ²-Bench Telecom (AA) and how often is it updated?
τ²-Bench Telecom (AA) is published by Artificial Analysis. BenchLeader re-reads the published results once a day and records the date each result was published.
Does τ²-Bench Telecom (AA) count toward the BenchLeader Index?
No. τ²-Bench Telecom (AA) is shown for reference but left out of the composite index.