VDiff-Bench: leaderboard
Metric: Choice Accuracy (%). Source: arxiv.org. 11 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 89.6 |
| 2 | Kimi K2.5 | 89.2 |
| 3 | Gemini 3.5 Flash | 88.8 |
| 4 | GPT-5.4 | 84.4 |
| 5 | Kimi K3 | 84.2 |
| 6 | Grok 4.3 | 67.3 |
| 7 | Gemini 2.5 Flash | 65.9 |
| 8 | Qwen 3 VL 8B (Thinking) | 58 |
Interactive version: theaggregate.ai/benchmark?slug=vdiff-bench · How It Works · Data refreshed daily, snapshot 2026-09-20.