AI models
τ-bench Verified (airline): AI model scores
130 AI models on τ-bench Verified (airline), per OpenRouter Benchmarks. Low discrimination: most models tie with the top score.
τ-bench Verified (airline), on the OpenRouter Benchmarks leaderboard (CC BY 4.0, latest version we checked), simulates an airline support agent that uses systems and must follow company policy until the end of the conversation. The score is the share of tasks completed within the rules. Here are the 130 models in this section with a score on τ-bench Verified (airline), per OpenRouter Benchmarks. Most are in a statistical tie with the top score, so the table shows the score without a rank or a badge.
τ-bench Verified (airline)
82%
Claude Fable 5
models
130
01
| Model | % | Source |
|---|---|---|
| Claude Fable 5Anthropic · closed | 82% | OpenRouter Benchmarks |
| DeepSeek V4 Flash 0423DeepSeek · open | 81.7% | OpenRouter Benchmarks |
| Gemini 3.7 FlashGoogle · closed | 80.7% | OpenRouter Benchmarks |
| Qwen3.5 397B A17BAlibaba · open | 80.2% | OpenRouter Benchmarks |
| GPT-5.1OpenAI · closed | 80% | OpenRouter Benchmarks |
| Ling 3.1 FlashAnt Group · undisclosed | 79.3% | OpenRouter Benchmarks |
| Claude Fable 5.1Anthropic · closed | 79.2% | OpenRouter Benchmarks |
| Qwen3.6 27BAlibaba · open | 79% | OpenRouter Benchmarks |
| MiMo V2.6 ProXiaomi · open | 78.9% | OpenRouter Benchmarks |
| Claude Sonnet 5.5Anthropic · closed | 78.8% | OpenRouter Benchmarks |
| GLM-5.3Zhipu AI (Z.ai) · open | 78.8% | OpenRouter Benchmarks |
| Claude Opus 5Anthropic · closed | 78.7% | OpenRouter Benchmarks |
| Gemini 4 ArgonGoogle · closed | 78.7% | OpenRouter Benchmarks |
| Kimi K3.1 ProMoonshot AI · open | 78.7% | OpenRouter Benchmarks |
| Qwen3.5-35B-A3BAlibaba · open | 78.7% | OpenRouter Benchmarks |
| DeepSeek V4 ProDeepSeek · open | 78.6% | OpenRouter Benchmarks |
| Gemini 3.5 Flash LiteGoogle · closed | 78.3% | OpenRouter Benchmarks |
| MiMo-V2.6-FlashXiaomi · open | 78.2% | OpenRouter Benchmarks |
| Qwen3.8 27BAlibaba · open | 78% | OpenRouter Benchmarks |
| Claude Opus 4.7Anthropic · closed | 77.8% | OpenRouter Benchmarks |
| DeepSeek V4.1 FlashDeepSeek · open | 77.8% | OpenRouter Benchmarks |
| Qwen3.5-122B-A10BAlibaba · open | 77.7% | OpenRouter Benchmarks |
| Hy4Tencent · open | 77.6% | OpenRouter Benchmarks |
| Claude Opus 4.8Anthropic · closed | 77.5% | OpenRouter Benchmarks |
| Qwen3.8 2.4T A95BAlibaba · open | 77.4% | OpenRouter Benchmarks |
| Kimi K2.7 CodeMoonshot AI · open | 77.3% | OpenRouter Benchmarks |
| Muse Glimmer 30BMeta · open | 77.2% | OpenRouter Benchmarks |
| Gemini 3.1 ProGoogle · closed | 77% | OpenRouter Benchmarks |
| Gemma 4 31BGoogle · open | 77% | OpenRouter Benchmarks |
| GLM 5Zhipu AI (Z.ai) · open | 76.9% | OpenRouter Benchmarks |
| DeepSeek V4 Flash Vision ExpDeepSeek · open | 76.7% | OpenRouter Benchmarks |
| Gemini 3.8 FlashGoogle · closed | 76.7% | OpenRouter Benchmarks |
| GLM-5.2Zhipu AI (Z.ai) · open | 76.7% | OpenRouter Benchmarks |
| Nova Micro 1.0Amazon · undisclosed | 76.7% | OpenRouter Benchmarks |
| Claude Opus 4.5Anthropic · closed | 76.6% | OpenRouter Benchmarks |
| Qwen3.8 FlashAlibaba · open | 76.4% | OpenRouter Benchmarks |
| Gemini 3.6 FlashGoogle · closed | 76.3% | OpenRouter Benchmarks |
| GLM-5.3-FlashZhipu AI (Z.ai) · open | 76.3% | OpenRouter Benchmarks |
| MiniMax M3MiniMax · open | 76.3% | OpenRouter Benchmarks |
| Kimi K3Moonshot AI · open | 76.2% | OpenRouter Benchmarks |
| Qwen3.6 35B A3BAlibaba · open | 76.2% | OpenRouter Benchmarks |
| GLM 4.7Zhipu AI (Z.ai) · open | 76% | OpenRouter Benchmarks |
| Claude Opus 4.6Anthropic · closed | 75.8% | OpenRouter Benchmarks |
| GLM 5.1Zhipu AI (Z.ai) · open | 75.8% | OpenRouter Benchmarks |
| Muse Spark OpenMeta · open | 75.7% | OpenRouter Benchmarks |
| Qwen3.5-9BAlibaba · open | 75.7% | OpenRouter Benchmarks |
| Claude Sonnet 5Anthropic · closed | 75.4% | OpenRouter Benchmarks |
| Claude Opus 5.5Anthropic · closed | 75.3% | OpenRouter Benchmarks |
| GPT-5.3-CodexOpenAI · closed | 75.3% | OpenRouter Benchmarks |
| Grok 4.6xAI · closed | 75.3% | OpenRouter Benchmarks |
| Grok 4.7xAI · closed | 75.3% | OpenRouter Benchmarks |
| GPT-5.4OpenAI · closed | 75% | OpenRouter Benchmarks |
| Claude Sonnet 4.5Anthropic · closed | 74.9% | OpenRouter Benchmarks |
| Kimi K2.6Moonshot AI · open | 74.9% | OpenRouter Benchmarks |
| Nemotron 3.5 Super VLNVIDIA · open | 74.9% | OpenRouter Benchmarks |
| Claude Sonnet 4.6Anthropic · closed | 74.8% | OpenRouter Benchmarks |
| DeepSeek V3.2DeepSeek · open | 74.7% | OpenRouter Benchmarks |
| GPT-5.5OpenAI · closed | 74.7% | OpenRouter Benchmarks |
| MiMo-V2.5Xiaomi · open | 74.7% | OpenRouter Benchmarks |
| MiniMax M3.1 FlashMiniMax · open | 74.7% | OpenRouter Benchmarks |
| GPT-5.6 SolOpenAI · closed | 74.2% | OpenRouter Benchmarks |
| MiMo-V2.5-ProXiaomi · open | 74.1% | OpenRouter Benchmarks |
| Claude Haiku 5.5Anthropic · closed | 74% | OpenRouter Benchmarks |
| Ling 3.0 FlashAnt Group · open | 74% | OpenRouter Benchmarks |
| Kimi K2.5Moonshot AI · open | 73.8% | OpenRouter Benchmarks |
| MiniMax M2.7MiniMax · open | 73.4% | OpenRouter Benchmarks |
| Gemini 3 Flash PreviewGoogle · undisclosed | 73.3% | OpenRouter Benchmarks |
| GLM 4.6Zhipu AI (Z.ai) · open | 73.3% | OpenRouter Benchmarks |
| DeepSeek V3.1DeepSeek · open | 73.1% | OpenRouter Benchmarks |
| DeepSeek V3.1 TerminusDeepSeek · open | 73% | OpenRouter Benchmarks |
| Gemini 3.5 FlashGoogle · closed | 73% | OpenRouter Benchmarks |
| Gemini 3.1 Flash LiteGoogle · closed | 72.7% | OpenRouter Benchmarks |
| GPT-5.6 LunaOpenAI · closed | 72.7% | OpenRouter Benchmarks |
| GPT-6 Astra ProOpenAI · undisclosed | 72.7% | OpenRouter Benchmarks |
| Qwen3.7 PlusAlibaba · closed | 72.7% | OpenRouter Benchmarks |
| GPT-5 MiniOpenAI · closed | 72.4% | OpenRouter Benchmarks |
| GPT-5.6 TerraOpenAI · closed | 72.3% | OpenRouter Benchmarks |
| Ling 3.0 Flash VLAnt Group · open | 72% | OpenRouter Benchmarks |
| Qwen3.7 MaxAlibaba · closed | 72% | OpenRouter Benchmarks |
| GPT-5.6 Sol ProOpenAI · undisclosed | 71.7% | OpenRouter Benchmarks |
| Nemotron 3.5 LightningNVIDIA · open | 71.5% | OpenRouter Benchmarks |
| GPT-5OpenAI · closed | 70.7% | OpenRouter Benchmarks |
| GPT-5.6 Luna ProOpenAI · undisclosed | 70.7% | OpenRouter Benchmarks |
| GPT-6.1 SolOpenAI · closed | 70.7% | OpenRouter Benchmarks |
| Gemma 4 26B A4BGoogle · open | 70.6% | OpenRouter Benchmarks |
| GPT-6 SolOpenAI · closed | 70.2% | OpenRouter Benchmarks |
| GPT-6.1 TerraOpenAI · closed | 70% | OpenRouter Benchmarks |
| GLM 4.7 FlashZhipu AI (Z.ai) · open | 69.7% | OpenRouter Benchmarks |
| GPT-6 AstraOpenAI · closed | 69.7% | OpenRouter Benchmarks |
| GPT-6 Luna ProOpenAI · undisclosed | 69.3% | OpenRouter Benchmarks |
| MiniMax M2.5MiniMax · open | 69.2% | OpenRouter Benchmarks |
| GPT-6 LunaOpenAI · closed | 68.4% | OpenRouter Benchmarks |
| gpt-oss-120bOpenAI · open | 67.3% | OpenRouter Benchmarks |
| GPT-5.4 MiniOpenAI · closed | 66.7% | OpenRouter Benchmarks |
| Claude Haiku 4.5Anthropic · closed | 66.4% | OpenRouter Benchmarks |
| GPT-5.4 NanoOpenAI · closed | 65.6% | OpenRouter Benchmarks |
| Granite 4.2 8BIBM · open | 64% | OpenRouter Benchmarks |
| Qwen3 235B A22B Thinking 2507Alibaba · open | 61.7% | OpenRouter Benchmarks |
| GLM 4.6VZhipu AI (Z.ai) · open | 60.7% | OpenRouter Benchmarks |
| Qwen3 Coder NextAlibaba · open | 60.7% | OpenRouter Benchmarks |
| Gemini 2.5 ProGoogle · closed | 60% | OpenRouter Benchmarks |
| Gemini 2.5 FlashGoogle · closed | 59.1% | OpenRouter Benchmarks |
| DeepSeek R1DeepSeek · open | 58% | OpenRouter Benchmarks |
| Qwen3 32BAlibaba · open | 54.3% | OpenRouter Benchmarks |
| Gemini 2.5 Flash LiteGoogle · undisclosed | 53.7% | OpenRouter Benchmarks |
| gpt-oss-20bOpenAI · open | 53.5% | OpenRouter Benchmarks |
| GPT-4.1OpenAI · closed | 52.3% | OpenRouter Benchmarks |
| Nemotron 3 Nano 30B A3BNVIDIA · open | 52% | OpenRouter Benchmarks |
| GPT-5 NanoOpenAI · closed | 48.3% | OpenRouter Benchmarks |
| Gemma 3 12BGoogle · open | 48% | OpenRouter Benchmarks |
| Qwen3 VL 235B A22B InstructAlibaba · open | 48% | OpenRouter Benchmarks |
| Qwen3-CoderAlibaba · open | 48% | OpenRouter Benchmarks |
| Qwen3-235B-A22BAlibaba · open | 47.9% | OpenRouter Benchmarks |
| Qwen3 Next 80B A3B InstructAlibaba · open | 47.5% | OpenRouter Benchmarks |
| Qwen3 30B A3BAlibaba · open | 47% | OpenRouter Benchmarks |
| Mistral Small 4Mistral AI · open | 46% | OpenRouter Benchmarks |
| Qwen3 Coder 30B A3B InstructAlibaba · open | 46% | OpenRouter Benchmarks |
| GPT-4.1 MiniOpenAI · closed | 45.1% | OpenRouter Benchmarks |
| Qwen3 14BAlibaba · open | 44% | OpenRouter Benchmarks |
| DeepSeek V3DeepSeek · open | 43.8% | OpenRouter Benchmarks |
| Llama 4 MaverickMeta · open | 43.4% | OpenRouter Benchmarks |
| GPT-4oOpenAI · closed | 43.2% | OpenRouter Benchmarks |
| Qwen3 30B A3B Instruct 2507Alibaba · open | 41.5% | OpenRouter Benchmarks |
| Qwen3 VL 30B A3B InstructAlibaba · open | 41.3% | OpenRouter Benchmarks |
| Llama 3.3 70B InstructMeta · open | 39.9% | OpenRouter Benchmarks |
| Qwen3 VL 8B InstructAlibaba · open | 32.7% | OpenRouter Benchmarks |
| GPT-4o-miniOpenAI · closed | 30.7% | OpenRouter Benchmarks |
| Qwen2.5 72B InstructAlibaba · open | 27.3% | OpenRouter Benchmarks |
| Llama 3.1 8B InstructMeta · open | 23.5% | OpenRouter Benchmarks |
| GPT-4.1 NanoOpenAI · closed | 14% | OpenRouter Benchmarks |
FAQ
Which model has the top score on τ-bench Verified (airline)?
The top score on τ-bench Verified (airline) belongs to Claude Fable 5, at 82% per OpenRouter Benchmarks. Most models are in a statistical tie with it, so the table shows the score without a rank or a badge.
What does τ-bench Verified (airline) measure?
τ-bench Verified (airline), on the OpenRouter Benchmarks leaderboard (CC BY 4.0, latest version we checked), simulates an airline support agent that uses systems and must follow company policy until the end of the conversation. The score is the share of tasks completed within the rules.
Which models are in this ranking?
The 130 models in this section with a published score on τ-bench Verified (airline), per OpenRouter Benchmarks. No version inherits another's score.
How to read
- Benchmark scores are percentages, and higher is better.
- Rank is shown as a range, «1st of 130». Within the range, the gap to neighboring models falls within the benchmark's margin. The «top score» badge marks the model with the benchmark's top score, and the «statistical tie with the top score» badge marks models whose gap to it is within the margin.
- On a saturated benchmark, the table shows the scores with the badge and without a rank order.
Sources
- Epoch AI, CC BY 4.0, latest version we checked.
- OpenRouter Benchmarks, CC BY 4.0, latest version we checked.
- OpenRouter, dados públicos da API; exports de benchmarks e rankings em CC BY 4.0, latest version we checked.
Dates come from the sources: OpenRouter listing or Hugging Face repo creation, which may differ from the official launch. Family and version were grouped by rule from the name.
How to cite
If you use this data, cite the page and the date you accessed it.
Adriano Meirinho. "τ-bench Verified (airline): AI model scores". meirinho.com.br. https://meirinho.com.br/en/ia/modelos/benchmark/tau-bench-airline/. Accessed November 21, 2026.
Third-party data follows the license of each source. The full table is available as CSV and JSON.