CAREBench - Negative Emotion Labels: leaderboard
Cognitive appraisal of emotion (CAREBench, Sun et al., 2026): the model reads a first-person narrative of a recent emotional event and answers as the narrator would, zero-shot, against the annotations the narrator gave for 1,000 narratives. The model selects every negative emotion felt from a fixed list (angry, worried, sad, ashamed and others); example-F1 in percent, higher is better.
Metric: Example-F1 (%; per-narrative F1 of the selected negative emotion labels against the narrator's own labels; 1,000 narratives, zero-shot, mean of five runs for open models). Source: arxiv.org. Saturation forecast: Around 2029. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 (Non-reasoning) | 63.9 | #158 (Claude Sonnet 4.6) |
| 2 | GPT-5.2 (Non-reasoning) | 63 | #172 (GPT-5.2) |
| 3 | Qwen 3.5 9B (Non-reasoning) | 61.7 | #803 (Qwen 3.5 9B) |
| 4 | Llama 3.1 8B Instruct | 60.1 | #4277 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=carebench-negative-emotion-labels · How It Works · Data refreshed daily, snapshot 2026-11-12.