OmniStarPro-RNG (Offline): leaderboard
Real-time narration generation on OmniStarPro (Yang et al., 2026), a benchmark of YouTube streams in 15 scenarios, under offline evaluation: the model narrates at prescribed timestamps, and a GPT-4o judge scores each narration against the ground-truth clip caption for semantic accuracy, language quality and completeness. Mean score on a 0-10 scale over 1,000 test streams; higher is better.
Metric: Semantic correctness (0-10; GPT-4o judge score of each narration against the ground-truth clip caption, mean of semantic accuracy, language quality and information completeness; real-time narration generation with responses decoded at prescribed timestamps, 1,000 evaluation streams of the OmniStarPro-Live partition). Source: arxiv.org. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 5.03 | #741 |
| 2 | MiniCPM-V-2.6 | 4.34 | #3095 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=omnistarpro-rng-offline · How It Works · Data refreshed daily, snapshot 2026-11-12.