MultiCaption (Zero-Shot): leaderboard

Metric: Accuracy (%; balanced test split, 4,920 claim pairs in 52 languages, claims in their original languages with the languages named in the prompt; zero-shot yes/no contradiction judgment at temperature 0.1, mean of five runs). Source: arxiv.org. Saturation forecast: Estimated already saturated. 5 models tracked.

Top models

#ModelScore
1Mistral 7B Instruct (v0.3)56
2Qwen 2.5 7B Instruct53.9
3Llama 3.1 8B Instruct52.4
4gemma-7B49.1
5Phi-4 Mini Instruct48.7

Interactive version: theaggregate.ai/benchmark?slug=multicaption-zero-shot · How It Works · Data refreshed daily, snapshot 2026-09-26.