Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

τ²-Bench Telecom

A dual-control conversational AI benchmark simulating technical support scenarios where both agent and user must coordinate actions to resolve telecom service issues. The benchmark introduces a new paradigm for evaluating conversational AI by simulating both the agent and user to actively modify a shared world state.
Source:

Model Performance

#1
Grok-3-Mini-Beta
90.0%
#2
GPT-5-Thinking
85.0%
#3
OpenAI-O3-medium
81.0%
#4
OpenAI-O3-high
81.0%
#5
Grok-4-Thinking
75.0%
#6
Claude-4.0-Opus-Thinking
71.0%
#7
Claude-4.1-Opus-Thinking
71.0%
#8
GPT-5-mini-Thinking
68.0%
#9
GPT-OSS-120B
66.0%
#10
Claude-4.0-Sonnet-Thinking
65.0%
#11
OpenAI-O1-1217
63.0%
#12
Kimi-K2-Instruct
61.0%
#13
OpenAI-O4-mini-high
56.0%
#14
OpenAI-O4-mini-medium
56.0%
#15
Claude-3.7-Sonnet-Thinking
55.0%
#16
Gemini-2.5-Pro-Thinking
54.0%
#17
Qwen-3-Thinking
53.0%
#18
Claude-4.0-Sonnet
52.0%
#19
Grok-3-Beta
49.0%
#20
GPT-4.1
47.0%
#21
DeepSeek-V3-0324
47.0%
#22
Mistral-Medium-3.1
41.0%
#23
DeepSeek-R1
37.0%
#24
GPT-5-nano-Thinking
36.0%
#25
Qwen-3
33.0%
#26
Qwen-3-Max-Preview
33.0%
#27
Gemini-2.5-Flash-Thinking
32.0%
#28
OpenAI-O3-mini-high
31.0%
#29
Mistral-Large-2
31.0%
#30
OpenAI-O3-mini-medium
29.0%
#31
Llama-3.3-70B
27.0%
#32
DeepSeek-V3
23.0%
#33
Llama-3.1-405B
19.0%
#34
Llama-4-Maverick-17B
18.0%
#35
Gemini-2.5-Flash
15.0%