ACEBench
Harbor: pendingACEBench is a comprehensive benchmark for evaluating LLM tool-usage ability.
- Samples
- 1,023
- Released
- 2025
- Scores
- 0
How to run
ltb · tau2-bench
$ npm i -g lettokenburn-cli$ ltb bench pull tau2-bench
External harness. pip install inspect-evals && inspect eval inspect_evals/tau2_airline --model <provider/model> (other tasks: tau2_banking, tau2_retail, tau2_telecom) https://github.com/UKGovernmentBEIS/inspect_evals
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
6 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Claude Opus 4.6 | 91.9 | N/A | Self-reported | low | Jun 1, 2026 | github.com |
| Claude Sonnet 4.6 | 91.7 | N/A | Self-reported | low | Jun 1, 2026 | github.com |
| GLM-5 | 89.7 | N/A | Self-reported | low | Jun 1, 2026 | github.com |
| Qwen 3.5 | 86.7 | N/A | Self-reported | low | Jun 1, 2026 | github.com |
| Gemini 3.1 Pro | 85.3 | N/A | Self-reported | low | Jun 1, 2026 | github.com |
| Gemini 3 Pro | 65.8 | N/A | Self-reported | low | Jun 1, 2026 | github.com |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsLoading community runs…
About
τ²-bench (Tau Benchmark) — 工具增强NLP任务（Pass^N）. Scoring: Pass@N. Difficulty: 极高. Imported from benchmark_research.md (2026-06).