FluVid (Per-Frame Input): leaderboard
Metric: Spearman rank correlation (SRCC, -1 to 1) between the model's predicted fluency score and the human mean opinion score (1 to 5) of the FluVid videos (real-world clips from SSv2 and five user-generated video-quality datasets, rated by 20 trained annotators for perceptual fluency, the smoothness of motion and playback), zero-shot: the model is asked 'How would you rate the fluency of this video?' and its probabilities over the words bad, poor, fair, good and excellent are turned into a 1-5 score, scoring 64 uniformly sampled frames one at a time with an image model and averaging the frame scores; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Q-Align | 0.57 | |
| 2 | Q-Align-VQA | 0.53 | |
| 3 | Q-Align-IQA | 0.5 | |
| 4 | Q-Align-IAA | 0.49 | |
| 5 | Q-Instruct | 0.39 | |
| 6 | LLaVA-OneVision-7B | 0.34 | |
| 7 | LLaVA-v1.5-7B | 0.3 | |
| 8 | ShareGPT4V-7B | 0.27 |
Interactive version: theaggregate.ai/benchmark?slug=fluvid-per-frame-input · How It Works · Data refreshed daily, snapshot 2026-10-11.