ViSTR-Bench: leaderboard
Metric: Accuracy (%; averaged over all 1,340 two-option questions in 15 video subtasks across motion perception, spatial relations, outcome prediction and physical dynamics; chance 50, majority-class 57.9). Source: arxiv.org. Saturation forecast: Around 2032. 39 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 (Thinking) | 62 |
| 2 | Seed 2.0 Pro (Thinking) | 60.1 |
| 3 | Seed 2.0 Pro (Non-reasoning) | 56.8 |
| 4 | GPT-5.4 (Non-reasoning) | 56.1 |
| 5 | Claude Opus 4.6 | 55.4 |
| 6 | Qwen 3.5 27B | 55 |
| 7 | Claude Opus 4.6 (Thinking) | 54.7 |
| 8 | Qwen 3.5 397B A17B | 54.4 |
| 9 | Gemini 3.1 Pro (Preview) | 53.6 |
| 10 | Qwen 3.5 122B A10B | 53.4 |
| 11 | Claude Sonnet 4.6 (Thinking) | 53.4 |
| 12 | Gemini 3.1 Flash Lite (Preview) | 52.6 |
| 13 | Intern-S1 | 51.7 |
| 14 | Claude Sonnet 4.6 | 51 |
| 15 | InternVL3.5-8B | 50.9 |
Interactive version: theaggregate.ai/benchmark?slug=vistr-bench · How It Works · Data refreshed daily, snapshot 2026-09-29.