CoWorld Reasoner: Agentic Multimodal Reasoning with the Evolving Video World Model-Driven Harness
CoWorld Reasoner lets a VLM call a video world model as a tool, lifting accuracy on all seven reasoning benchmarks.
CoWorld Reasoner is an agentic framework that lets a vision-language model decide when and how to invoke a video world model for visual imagination. A harness exposes visual operations as tools and skills, and a co-evolution loop revises tools, skills, and execution code from feedback while keeping both model weights fixed. Across seven benchmarks covering spatial, temporal, causal, and perceptual reasoning, the strongest variant improves every benchmark, including relative accuracy gains of 11.8% on MMSI and 13.0% on ERQA. It also outperforms five evolving baselines that use the same backbone on five benchmarks.
- Uses a video world model as a tool-driven visual imagination engine.
- Harness co-evolution revises tools, skills, and code with model weights fixed.
- Strongest variant improves all seven spatial, temporal, causal, and perceptual benchmarks.
- Relative gains include 11.8% on MMSI and 13.0% on ERQA.
- Beats five evolving baselines with the same backbone on five benchmarks.
- AI models
- CoWorld Reasoner
From the original article
1 min read at huggingface.co
As Vision-Language Models (VLMs) advance, they are increasingly expected to perform spatiotemporal reasoning over motion, actions, and view changes beyond what the input images show. Video World Models (VWMs) provide a natural unified engine for visual imagination by generating task-relevant visual content as additional evidence.
By Dongjie Cheng, Ruifeng Yuan, Ran Wei · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.