CAREBench - Positive Emotion Labels: leaderboard

Metric: Example-F1 (%; per-narrative F1 of the selected positive 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 2031. 6 models tracked.

Top models

#ModelScore
1GPT-5.2 (Non-reasoning)55.6
2Qwen 3.5 9B (Non-reasoning)51.3
3Claude Sonnet 4.6 (Non-reasoning)49
4Llama 3.1 8B Instruct45

Interactive version: theaggregate.ai/benchmark?slug=carebench-positive-emotion-labels · How It Works · Data refreshed daily, snapshot 2026-09-26.