{"slug": "vgi-bench-probing-visual-intelligence-in-video-generation-models", "title": "VGI-BENCH: Probing Visual Intelligence in Video Generation Models", "summary": "Researchers introduced VGI-bench, a benchmark with 27 tasks and 810 instances for evaluating visual reasoning in video generation models, finding that even the strongest model, Seedance 2.0, achieves only 51.0% accuracy. The benchmark reveals current systems solve a subset of visually grounded reasoning tasks but remain unreliable, with limited self-correction during denoising. The team will release code and data to stimulate development of next-generation video generation models.", "body_md": "arXiv:2608.19583v1 Announce Type: new\nAbstract: Recent studies suggest that video generation models can exhibit certain forms of zero-shot visual reasoning through generated frames. Yet reliable evaluation remains challenging: benchmarks should adopt inputs aligned with the visual priors of current video models, require valid evolving processes rather than only plausible final states, and calibrate task difficulty to remain challenging yet partly feasible. To this end, we introduce VGI-bench, containing 27 tasks and 810 instances, organized by a two-level taxonomy of task domains and skill tags for fine-grained evaluation of visual reasoning capabilities of video generation models. Our evaluations show that current generative systems can solve a subset of visually grounded reasoning tasks, but remain far from reliable, with even the strongest model, Seedance~2.0, achieving only 51.0% under our evaluation criteria. Our analysis further explore the output failure modes, input condition sensitivity, performance transfer boundary from synthetic fine-tuning, and internal denoising perspective revealing limited self-correction, where later steps mainly refine early hypotheses rather than correct reasoning errors. We hope VGI-bench will help stimulate the development of next-generation video generation models. We will release our code and data.", "url": "https://wpnews.pro/news/vgi-bench-probing-visual-intelligence-in-video-generation-models", "canonical_source": "https://arxiv.org/abs/2608.19583", "published_at": "2026-08-21 04:00:00+00:00", "updated_at": "2026-08-21 04:16:58.185614+00:00", "lang": "en", "topics": ["artificial-intelligence", "computer-vision", "ai-research", "ai-tools"], "entities": ["VGI-bench", "Seedance 2.0", "arXiv"], "alternates": {"html": "https://wpnews.pro/news/vgi-bench-probing-visual-intelligence-in-video-generation-models", "markdown": "https://wpnews.pro/news/vgi-bench-probing-visual-intelligence-in-video-generation-models.md", "text": "https://wpnews.pro/news/vgi-bench-probing-visual-intelligence-in-video-generation-models.txt", "jsonld": "https://wpnews.pro/news/vgi-bench-probing-visual-intelligence-in-video-generation-models.jsonld"}}