MANTA - Moral Sensitivity: leaderboard

The moral sensitivity dimension of MANTA (2026): whether a model spontaneously and appropriately raises animal welfare stakes in a first-turn request that never mentions them, such as a procurement or recipe question. Judged on a 0-100 scale; higher is better.

Metric: Animal Welfare Moral Sensitivity (0-100): turn-1 rubric score for spontaneously and appropriately surfacing the welfare stakes of an implicitly framed request, 788 implicit-framing animal-welfare scenarios expanded to about 1,088 per model, one conversation each with an explicit welfare prompt at turn 2 and three adversarial pressure rounds, LLM judge (Claude Sonnet 4.6, GPT-5.4 for Claude Opus 4.7) on a 0-1 rubric, times 100; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 7 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Opus 4.757.9#85
2GPT-5.550.4#54
3Llama 3.3 70B Instruct47.6#1385
4DeepSeek V4 Flash41.7#267
5Gemini 3.1 Flash Lite40.1#371
6Grok 4.337.1#342
7Mistral Small 436.5#1229

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=manta-moral-sensitivity · How It Works · Data refreshed daily, snapshot 2026-11-21.