AttuneBench - Pairwise Accuracy: leaderboard
AttuneBench (2026) scores LLM emotional intelligence on genuine multi-turn conversations whose participants annotated their own emotions, preferred responses and judgments of how the model behaved. This board is the pairwise accuracy (%) with which a model predicts which of three responses per turn the participant preferred, in the default mode on 200 conversations; higher is better.
Metric: Pairwise accuracy (%) at predicting the participant's preferences among three response variants per turn (original, preference-informed alternate, participant-written golden), Default mode (conversation only), 200 conversations; higher is better. Source: arxiv.org. Saturation forecast: Around 2030. 11 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 64.6 | #86 |
| 2 | Claude Opus 4.6 | 63.7 | #114 |
| 3 | MiMo-V2-Pro | 59.8 | #356 |
| 4 | Claude Haiku 4.5 | 56 | #587 |
| 5 | GPT-5.5 | 53.8 | #56 |
| 6 | Mistral Large | 51.7 | #2144 |
| 7 | Claude Sonnet 4.6 | 51.2 | #179 |
| 8 | Gemini 3.1 Pro (Preview) | 50.3 | #102 |
| 9 | GPT-5.4 | 47 | #138 |
| 10 | Grok 4 | 46.1 | #321 |
| 11 | Qwen 2.5 72B | 45 | #1189 |
No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=attunebench-pairwise-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-22.