VLM-3R Release VSI-Bench - Overall: leaderboard
VLM3R release original v1: numeric tasks use mean relative accuracy, multiple-choice tasks useaccuracy. Full VSI population separate from TinySet. Human/chance rows are not model aliases. VSTI-specific adapter is not the ordinary VSI adapter. Metric: Source benchmark score (%). Higher is better. Source: https://arxiv.org/html/2505.20279v1
Metric: Source benchmark score (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 1.5 Pro | 45.4 | #443 |
| 2 | Gemini 1.5 Flash | 42.1 | #615 |
| 3 | InternVL2-8B | 34.6 | #813 |
| 4 | GPT-4o | 34 | #339 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #7 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=vlm-3r-release-vsi-bench-overall · How It Works · Data refreshed daily, snapshot 2026-10-17.