TAU-bench Retail

Agentic

Source ↗ Higher is better

Rank	Model	Provider	Score
1	Anthropic: Claude Sonnet 4.5	Anthropic	86.2%
2	Gemini 3.0 Pro	Google	85.3%
3	Qwen: Qwen3.5 397B A17B	Alibaba	84.4%
4	Nemotron-Orchestrator-8B	NVIDIA	84.2%
5	RAFT-30B-A3B	Northeastern University HAI Lab	82.5%
6	Anthropic: Claude Opus 4.1	Anthropic	82.4%
7	OpenAI: GPT-5.2	OpenAI	81.6%
8	OpenAI: GPT-5	OpenAI	81.6%
9	Claude Opus 4	Anthropic	81.4%
10	DeepSeek: DeepSeek V3.2Historical	DeepSeek	81.1%
11	Anthropic: Claude Sonnet 4Historical	Anthropic	80.5%
12	Anthropic: Claude Opus 4.5	Anthropic	79.6%
13	Qwen: Qwen3 Max ThinkingHistorical	Alibaba	79.4%
14	Gemini 3 Flash	Google	76.8%
15	gemini-3-pro	Google	75.9%
16	OpenAI: GPT-4.1	OpenAI	74.0%
17	OpenAI: o3	OpenAI	73.9%
18	Z.ai: GLM 5	Z.ai	73.7%
19	Qwen: Qwen3 MaxHistorical	Alibaba	72.2%
20	Claude-3.7-Sonnet	Anthropic	72.1%
21	MoonshotAI: Kimi K2 0711	Moonshot AI	70.6%
22	OpenAI: o4 Mini	OpenAI	68.3%
23	OpenAI: GPT-4.1 Mini	OpenAI	61.4%

Access this data

JSON GET /api/v1

[
  {
    "rank": 1,
    "model": "Anthropic: Claude Sonnet 4.5",
    "provider": "Anthropic",
    "score": "86.2"
  },
  {
    "rank": 2,
    "model": "Gemini 3.0 Pro",
    "provider": "Google",
    "score": "85.3"
  },
  {
    "rank": 3,
    "model": "Qwen: Qwen3.5 397B A17B",
    "provider": "Alibaba",
    "score": "84.43"
  }
]

Curl

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

CLI brew install rightrank

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

Claude Code Skill Skill

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