M2-Verify-Gen: leaderboard
M2-Verify (2026) is a multimodal scientific claim consistency benchmark: a model sees a figure, its caption and a claim and decides zero-shot whether the claim is supported or refuted. This board is macro-F1 in percent on the M2-Verify-Gen test set, figures and captions from 16 arXiv categories with negated claims; higher is better.
Metric: Macro-F1 (%) of binary claim consistency verification (supported or refuted) on the M2-Verify-Gen test set (figures and captions from 16 arXiv categories, negation-perturbed claims), zero-shot; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 7B Instruct | 68.1 | #1919 |
| 2 | InternVL3-8B | 63.8 | #1795 |
| 3 | Mistral Small 3.1 | 63.1 | #1734 |
| 4 | GPT-5 Mini | 59.9 | #373 |
| 5 | GPT-4o Mini | 56.6 | #1680 |
| 6 | Llama 3.2 11B Vision | 53.8 | #4281 |
| 7 | Phi-4 Multimodal Instruct | 45.7 | #4155 |
| 8 | Pixtral-12B | 41.9 | #2611 |
No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=m2-verify-gen · How It Works · Data refreshed daily, snapshot 2026-11-22.