AttuneBench (Verbose): leaderboard
Metric: Composite (0-100) = 100 x (0.24 emotion tracking + 0.49 turn-level verification + 0.27 holistic comprehension), the evaluated model stepping through real human-model conversations turn by turn against the participants' own annotations, Verbose mode (the model also writes out its reasoning), 50-conversation subsample; higher is better. Source: arxiv.org. Saturation forecast: Around 2032. 11 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.7 | 53.95 |
| 2 | Claude Opus 4.6 | 53.5 |
| 3 | GPT-5.5 | 53.44 |
| 4 | MiMo-V2-Pro | 52.71 |
| 5 | Qwen 2.5 72B | 51.48 |
| 6 | Claude Haiku 4.5 | 51.47 |
| 7 | Gemini 3.1 Pro (Preview) | 51.21 |
| 8 | Claude Sonnet 4.6 | 49.73 |
| 9 | Grok 4 | 49.7 |
| 10 | Mistral Large | 49.66 |
| 11 | GPT-5.4 | 49.1 |
Interactive version: theaggregate.ai/benchmark?slug=attunebench-verbose · How It Works · Data refreshed daily, snapshot 2026-10-07.