τ²-Bench Airline Domain (τ²-bench Airline)
We show this table for reference; we do not rank on it.
τ²-bench Airline tests conversational agents on airline customer-service tasks governed by domain policy and database-changing tools.
Benchmark score on τ²-bench Airline — October 28, 2026
We compile the τ²-bench Airline rows from provider self-reports. The table contains 5 models. Trinity-Large-Thinking has the highest published score at 88.0%, but the available coverage and evaluation setups do not establish a market leader. We do not use these results to rank models overall.
Trinity-Large-Thinking
Arcee AI
τ²-bench Airline
Nemotron 3.5 Super VL 120B A12B
NVIDIA
τ²-bench Airline
Nemotron 3.5 Super VL 120B A12B NVFP4
NVIDIA
τ²-bench Airline
5 modelsAgenticCurrentDisplay onlyUpdated October 28, 2026
Benchmark score table (5 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Trinity-Large-ThinkingArcee AIτ²-bench Airline | 88.0% | Not reported | Open |
| 2 | Nemotron 3.5 Super VL 120B A12BNVIDIAτ²-bench Airline | 76.5% | Not reported | Open |
| 3 | Nemotron 3.5 Super VL 120B A12B NVFP4NVIDIAτ²-bench Airline | 75.8% | Not reported | Open |
| 4 | Ember-1Fireworksτ²-bench Airline | 66.0% | Not reported | Closed |
| 5 | ZAYA1-74B-PreviewZyphraτ²-bench Airline · avg@1 | 56.1% | Not reported | Open |
How to read this leaderboard
Editorial review by Glevd · 2026-07-15
Treat the current display as one source receipt, not a market ranking. Compare another airline result only when the task release, agent and user models, scaffold, prompts, trial count, and avg@1 or pass^k policy match.
Operator receipt: 5 sourced rows are currently displayable on this page; the highest published score among these 5 models is Trinity-Large-Thinking at 88.0%, which does not establish a market leader.
Honest limit: Only one of six stored airline rows currently has an exact source attachment. The other five remain withheld. One sourced row cannot establish a leader, and a simulated airline workflow does not cover production authentication, permissions, live inventory, policy drift, or human escalation.
Among the reported τ²-bench Airline rows, Trinity-Large-Thinking is first at 88.0%. The third row is 12.3 points behind. The broader top-10 range is 31.9 points, so the table still separates the published systems.
5 models have been evaluated on τ²-bench Airline. The benchmark falls in the Agentic category. τ²-bench Airline is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About τ²-bench Airline
Year
2025
Tasks
Airline customer-service tasks
Format
Domain success under a published trial policy
Difficulty
Policy-constrained airline support workflows
This lane owns results explicitly labeled for the airline domain. It stays separate from telecom, retail, aggregates, the archived original TAU-bench release, and newer τ³-bench runs.
Freshness and provenance
Version
τ²-bench Airline 2025
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does τ²-bench Airline measure?
τ²-bench Airline tests conversational agents on airline customer-service tasks governed by domain policy and database-changing tools.
Which model scores highest on τ²-bench Airline?
Trinity-Large-Thinking by Arcee AI currently leads with a score of 88.0% on τ²-bench Airline.
How many models are evaluated on τ²-bench Airline?
5 AI models have published results on τ²-bench Airline in the BenchLM catalog.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.