Tau2 Telecom Benchmark Leaderboard

τ²-Bench telecom domain evaluates conversational agents in a dual-control environment modeled as a Dec-POMDP, where both agent and user use tools in shared telecommunications troubleshooting scenarios that test coordination and communication capabilities.

Leaderboard

Top 30 models on Tau2 Telecom Benchmark Leaderboard (scores from public evaluations).

Rank	Model	Score	Lab
1	LongCat-Flash-Thinking-2601	99.3%	—
1	Claude Opus 4.6	99.3%	—
3	GPT-5.4	98.9%	—
4	GPT-5.2	98.7%	—
5	Claude Opus 4.5	98.2%	—
6	GPT-5.5	98.0%	—
7	Claude Sonnet 4.6	97.9%	—
8	MiMo-V2-Pro	96.8%	—
9	GPT-5	96.7%	—
10	GPT-5.1	95.6%	—
10	GPT-5.1 Instant	95.6%	—
10	GPT-5.1 Thinking	95.6%	—
13	GPT-5.4 mini	93.4%	—
14	GPT-5.4 nano	92.5%	—
15	Muse Spark	91.5%	—
16	MiniMax M2	87.0%	—
16	MiniMax M2.1	87.0%	—
18	LongCat-Flash-Thinking	83.1%	—
19	Claude Haiku 4.5	83.0%	—
20	LongCat-Flash-Chat	73.7%	—
21	LongCat-Flash-Lite	72.8%	—
22	Kimi K2-Instruct-0905	65.8%	—
22	Kimi K2 Instruct	65.8%	—
24	Nemotron 3 Super (120B A12B)	64.4%	—
25	o3	58.2%	—
26	Qwen3-235B-A22B-Thinking-2507	45.6%	—
27	Qwen3-Next-80B-A3B-Thinking	43.9%	—
28	Nemotron 3 Nano (30B A3B)	42.2%	—
29	GPT-4o	23.5%	—
30	Qwen3-Next-80B-A3B-Instruct	13.2%	—

Models tracked

Models with tau2-telecom in their evaluation profile.

No models linked yet.

View task leaderboards →