evals.report
BenchmarksLabsCompareRun guidesIn the wild
BenchmarksTool use

τ²-bench (Telecom)

A dual-control, multi-turn tool-agent-user benchmark (telecom split) where both the AI agent and a simulated user invoke tools to coordinate and resolve technical-support troubleshooting tasks in a shared, dynamic environment.

Tool usepass^1Higher is better
ScoresAboutRun this benchmark

What is τ²-bench (Telecom)?

A dual-control, multi-turn tool-agent-user benchmark (telecom split) where both the AI agent and a simulated user invoke tools to coordinate and resolve technical-support troubleshooting tasks in a shared, dynamic environment. evals.report tracks reported τ²-bench (Telecom) scores with the model, source, status, date, and run caveats attached — official leaderboard scores, vendor-reported launches, and clearly labeled community runs.

Top reported τ²-bench (Telecom) score: Claude Opus 4.6 — 99.3% (pass^1).

ModelLabScore↓Source modelStatusDate
Claude Opus 4.6Anthropic99.3%—UnverifiedFeb 5, 2026Details
GPT-5.2OpenAI98.7%—UnverifiedDec 11, 2025Details
Claude Opus 4.5Anthropic98.2%—UnverifiedNov 24, 2025Details
Grok 4.3xAI97.7%—OfficialApr 17, 2026Details
GLM-5.1OpenZ.ai97.7%—OfficialApr 7, 2026Details
GPT-5OpenAI96.7%—UnverifiedAug 7, 2025Details
DeepSeek V4 ProOpenDeepSeek96.2%—OfficialApr 24, 2026Details
Kimi K2.6OpenMoonshot AI95.9%—OfficialApr 20, 2026Details
Gemini 3.1 Pro PreviewGoogle DeepMind95.6%—OfficialFeb 19, 2026Details
Qwen3.5-397B-A17BOpenAlibaba / Qwen95.6%—OfficialFeb 16, 2026Details
GPT-5.1OpenAI95.6%—UnverifiedNov 12, 2025Details
Gemini 3.5 FlashGoogle DeepMind95.3%—OfficialMay 19, 2026Details
DeepSeek V4 FlashOpenDeepSeek95.0%—OfficialApr 24, 2026Details
Qwen3.7 Max PreviewAlibaba / Qwen94.7%—OfficialMay 14, 2026Details
Claude Opus 4.8Anthropic94.4%—OfficialMay 28, 2026Details
Mistral Medium 3.5Mistral AI94.2%—OfficialApr 28, 2026Details
MiMo-V2.5OpenXiaomi94.2%—OfficialApr 22, 2026Details
GPT-5.5OpenAI93.9%—OfficialApr 23, 2026Details
Amazon Nova 2 ProAmazon92.7%—OfficialDec 2, 2025Details
Muse SparkMeta91.5%—OfficialApr 8, 2026Details
Claude Opus 4.7Anthropic88.6%—OfficialApr 16, 2026Details
GPT-5.4OpenAI87.1%—OfficialMar 5, 2026Details
MiniMax M2.1OpenMiniMax87.0%—UnverifiedDec 23, 2025Details
Gemini 3 ProGoogle DeepMind85.4%—UnverifiedNov 18, 2025Details
MiniMax M2.7OpenMiniMax84.8%—OfficialMar 18, 2026Details
Claude Sonnet 4.6Anthropic75.7%—OfficialFeb 17, 2026Details
NVIDIA Nemotron 3 Super 120B-A12BOpenNVIDIA67.8%—OfficialMar 10, 2026Details
GPT-OSS-120BOpenOpenAI65.8%—OfficialAug 5, 2025Details
Kimi K2 InstructOpenMoonshot AI65.8%—UnverifiedJul 11, 2025Details
o3OpenAI58.2%—UnverifiedApr 16, 2025Details
Claude Haiku 4.5Anthropic54.7%—OfficialOct 15, 2025Details
Claude 3.7 SonnetAnthropic49%—VerifiedFeb 24, 2025Details
o4-miniOpenAI42%—VerifiedApr 16, 2025Details
GPT-4.1OpenAI34%—VerifiedApr 14, 2025Details
GPT-4oOpenAI23.5%—UnverifiedMay 13, 2024Details

Each row reports the model’s pass^1 on τ²-bench (Telecom). Click a row for the full run context.