CompanionHarm (Zero-Shot): leaderboard

Metric: Macro-F1 (%; 14-way single-label classification of annotated Replika AI companion utterances into 13 socio-emotional harm categories or no harm, each with its preceding multi-turn context; temperature 0; zero-shot prompt with task instruction and label set). Source: arxiv.org. Saturation forecast: Around 2029. 7 models tracked.

Top models

#ModelScore
1GPT-5.544.4
2Gemini 3.1 Pro (Preview)44
3Claude Opus 4.742.2
4Llama 3.1 70B Instruct37.5
5Qwen 3 32B33.7
6Llama 3.1 8B Instruct17
7Qwen 3 8B15.5

Interactive version: theaggregate.ai/benchmark?slug=companionharm-zero-shot · How It Works · Data refreshed daily, snapshot 2026-09-29.