MLVU-Dev - Generation: leaderboard
MLVU Dev set generation average (G-Avg): sub-scene captioning and whole-video summarization of long videos, each answer scored by GPT-4 against human annotations on a 0-10 scale (accuracy and relevance for captions, completeness and reliability for summaries). Higher is better (BAAI, CVPR 2025).
Metric: Generation score, GPT-4 judged (0-10). Source: github.com. Saturation forecast: Estimated already saturated. 30 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 5.8 | #741 |
| 2 | GPT-4 Turbo | 5.35 | #1002 |
| 3 | Qwen-VL-Max | 3.96 | #1226 |
| 4 | Claude 3 Opus | 3.39 | #1459 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mlvu-dev-generation · How It Works · Data refreshed daily, snapshot 2026-11-12.