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
| # | Model | Score |
|---|---|---|
| 1 | Mistral 7B Instruct (v0.3) | 56 |
| 2 | Qwen 2.5 7B Instruct | 53.9 |
| 3 | Llama 3.1 8B Instruct | 52.4 |
| 4 | gemma-7B | 49.1 |
| 5 | Phi-4 Mini Instruct | 48.7 |
Interactive version: theaggregate.ai/benchmark?slug=multicaption-zero-shot · How It Works · Data refreshed daily, snapshot 2026-09-26.