TAU-bench Telecom

Agentic

Source ↗ Higher is better

Rank	Model	Provider	Score
1	Qwen: Qwen3 Max ThinkingHistorical	Alibaba	98.2%
2	Anthropic: Claude Sonnet 4.5	Anthropic	98.0%
3	Gemini 3.0 Pro	Google	98.0%
4	Qwen: Qwen3.5 397B A17B	Alibaba	97.8%
5	DeepSeek: DeepSeek V3.2Historical	DeepSeek	96.2%
6	OpenAI: GPT-5	OpenAI	95.8%
7	Anthropic: Claude Opus 4.5	Anthropic	92.3%
8	Gemini 3 Flash	Google	91.2%
9	gemini-3-pro	Google	91.0%
10	OpenAI: GPT-5.2	OpenAI	89.7%
11	Nemotron-Orchestrator-8B	NVIDIA	88.6%
12	Z.ai: GLM 5	Z.ai	86.8%
13	Qwen: Qwen3 MaxHistorical	Alibaba	84.2%
14	MoonshotAI: Kimi K2 0711	Moonshot AI	65.8%
15	OpenAI: o4 Mini	OpenAI	50.2%
16	Claude-3.7-Sonnet	Anthropic	49.0%
17	OpenAI: GPT-4.1 Mini	OpenAI	48.9%
18	OpenAI: GPT-4.1	OpenAI	34.0%

Access this data

JSON GET /api/v1

[
  {
    "rank": 1,
    "model": "Qwen: Qwen3 Max Thinking",
    "provider": "Alibaba",
    "score": "98.24561403508771"
  },
  {
    "rank": 2,
    "model": "Anthropic: Claude Sonnet 4.5",
    "provider": "Anthropic",
    "score": "98.0"
  },
  {
    "rank": 3,
    "model": "Gemini 3.0 Pro",
    "provider": "Google",
    "score": "98.0"
  }
]

Curl

$ curl -s https://rightrank.ai/api/v1/rankings/tau-bench-telecom.json | jq '.[:3]'

CLI brew install rightrank

$ rightrank top tau-bench-telecom --limit 5
$ rightrank compare claude-opus-4.6 gpt-5.4-pro --category tau-bench-telecom
$ rightrank value tau-bench-telecom  # best performance per dollar

Claude Code Skill Skill

# In Claude Code, use the RightRank skill:
/rightrank best model for tau-bench-telecom
/rightrank compare top 5 by value
/rightrank cheapest model scoring above 80 on tau-bench-telecom