AI benchmarks and leaderboards are different things
Learn the difference between a benchmark, an evaluation run and a leaderboard, and use each to answer the right question.
Model reports
tau2-bench
Per-domain scores (Retail, Telecom, Airline) diverge sharply and several cards apply the Airline domain fixes from the Claude Opus 4.5 system card, so an average conceals both the domain mix and the patch level.
Evaluation context and evidence
- Source
- Model reports
- Publisher
- Not recorded
- Benchmark release
- 2025-06-09
- Collected
- Not recorded
- Modality
- Not recorded
These are source-reported observations, not independent reruns by this site. Check the task version, evaluation protocol, unit and reporting date before comparing models. Results from different setups are not a universal ranking.
Original sources
Reported results 5
Shown in source observation order, not ranked. Dates retain the precision provided by the source.
| Model | Value | Unit | Protocol | Reported date and basis | Source |
|---|---|---|---|---|---|
| Gemini 3.1 Pro | 90.8 | percent | Thinking (High), Retail | 2026-02-19 · document_publication | Gemini 3.1 Pro |
| Gemini 3.1 Pro | 99.3 | percent | Thinking (High), Telecom | 2026-02-19 · document_publication | Gemini 3.1 Pro |
| Gemma 4 (31B) | 76.9 | percent | average over 3 | 2026-03-11 · document_publication | Gemma 4 (31B) |
| Qwen3.5-397B-A17B | 86.7 | percent | thinking (temp 0.6, top_p 0.95, top_k 20) | 2026-02-16 · document_publication | Qwen3.5-397B-A17B |
| GLM-5 | 89.7 | percent | Retail and Telecom prompt adjustment; Airline domain fixes per Claude Opus 4.5 system card | 2026-02-11 · document_publication | GLM-5 |