VSTAT: leaderboard
VSTAT (Visual State Tracking benchmark, 2026) asks multimodal models 1,500 questions about 834 synthetic and real-world video clips that can only be answered by tracking a visual state through the whole video. The score (%) averages multiple-choice accuracy and mean relative accuracy on numerical answers over all questions; higher is better.
Metric: Average score (%) over all questions: 1,500 questions on 834 synthetic (Blender) and real-world video clips that require tracking a visual state through the whole video; accuracy on multiple-choice questions and mean relative accuracy on numerical questions (%), open models at their best of 16 to 128 uniformly sampled frames; higher is better. Source: arxiv.org. Saturation forecast: Around 2030. 23 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) (Low) | 44.4 | #94 (Gemini 3.1 Pro (Preview)) |
| 2 | Gemini 3.1 Pro (Preview) (High) | 43.9 | #94 (Gemini 3.1 Pro (Preview)) |
| 3 | Gemini 3 Flash (Low) | 39.8 | #164 (Gemini 3 Flash) |
| 4 | Gemini 3 Flash (High) | 38.8 | #164 (Gemini 3 Flash) |
| 5 | Molmo2-8B | 34 | #1483 |
| 6 | Qwen 3 VL 8B Instruct | 33.2 | #981 |
| 7 | Qwen 3 VL 4B Instruct | 31.3 | #1279 |
| 8 | GLM-4.1V-9B (Thinking) | 30.2 | #1060 (GLM-4.1V-9B) |
| 9 | Qwen 3 VL 8B (Thinking) | 28.2 | #887 |
| 10 | Qwen 3 VL 4B (Thinking) | 26 | #1296 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=vstat · How It Works · Data refreshed daily, snapshot 2026-11-13.