FewMMBench (4-Shot, Similar): leaderboard
Metric: Mean accuracy (%) on the FewMMBench query sets (about 250 multiple-choice or caption/foil items per skill with 2-4 options, curated by graph-cut selection from MMBench, SeedBench, ARO, VALSE, VSR, What'sUp, GVTBench, MME, CV-Bench and MileBench); regex answer extraction, four demonstrations selected by visual and textual similarity; the unweighted mean of the nine skill accuracies; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 7B Instruct | 83.9 | #643 |
| 2 | InternVL3-8B | 82.4 | #606 |
| 3 | Qwen 2 VL 7B Instruct | 80.3 | #816 |
| 4 | Phi-4 Multimodal Instruct | 67.4 | #896 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=fewmmbench-4-shot-similar · How It Works · Data refreshed daily, snapshot 2026-10-11.