Research map / Video / Video-language understanding
How Far Are Video Models from True Multimodal Reasoning?
Xiaotian Zhang, Jianhui Wei, Yuan Wang, Jie Tan, Yichen Li, Yan Zhang, Ziyi Chen, Daoan Zhang, DEZHI YU, Wei Xu, Songtao Jiang, Zuozhu Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add TextGrad prompt optimization (ablation) | Seed 2.0 Lite judge · AVE Perception | MCC: 43.3 -> 67.1 | Improved |
| add SemanticMatch on top of TextGrad (ablation) | Seed 2.0 Lite judge · AVE P.S.+L.R. | MCC: 48.5 -> 57.7 | Improved |
| add SemanticMatch on top of GEPA (ablation) | Seed 2.0 Lite judge · AVE E.E.+P.R.+S.C. | MCC: 39.3 -> 47.8 | Improved |
| add SemanticMatch on top of GEPA (ablation) | Seed 2.0 Lite judge · AVE P.S.+L.R. | MCC: 54.6 -> 51.6 | Hurt |
| add SemanticMatch to APO (ablation) | Seed 2.0 Lite judge · AVE, all tasks | MCC: avg +4.9% MCC; one of six TextGrad/GEPA task combos negative | Mixed |
| transfer prompt optimized on Seed 2.0 Lite to Seed 2.0 Pro | Seed 2.0 Pro judge · AVE three tasks | MCC: avg +13.2% MCC | Improved |
| prompt optimization of judge (ablation) | Seed 2.0 Lite judge · Kendall tau, Perception | Kendall tau: 0.433 -> 0.707 | Improved |
| VLM context-understanding rewrite of instruction (ablation) | Seedance 1.5 · CLVG-Bench Physical Simulation | pass rate: 37.5 -> 52.4 | Improved |
| VLM rewrite of instruction (ablation) | Seedance 2.0 · CLVG-Bench Logical Reasoning | pass rate: 21.3 -> 49.0 | Improved |
| VLM assistance in multi-turn interaction (ablation) | top video models · CLVG-Bench multi-turn | first-turn success: +44% and +50% | Improved |
Closest papers by meaning
- SciVideoBench: Benchmarking Scientific Video Reasoning in Large Multimodal Models (ICML 2026 · similarity 0.92)
- Thinking with Video: Video Generation as a Promising Multimodal Reasoning Paradigm (CVPR 2026 · similarity 0.90)
- MMR-V: What's Left Unsaid? A Benchmark for Multimodal Deep Reasoning in Videos (ICLR 2026 · similarity 0.90)
- RPTS: Tree-Structured Reasoning Process Scoring for Faithful Multimodal Evaluation (AAAI 2026 · similarity 0.90)
- VideoReasonBench: Can MLLMs Perform Vision-Centric Complex Video Reasoning? (ICLR 2026 · similarity 0.90)
- OneThinker: All-in-one Reasoning Model for Image and Video (CVPR 2026 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
