NVIDIA Releases Cosmos 3: A Two-Tower Mixture-of-Transformers Foundation Model Unifying Physical Reasoning, World Generation, and Action Generation
NVIDIA just open-sourced Cosmos 3. A dual-tower world model that reasons about physics, generates environments, and outputs actions. Here's why robotics teams should care.
Why it matters
NVIDIA's Cosmos 3 represents a significant architectural shift in foundation models—moving beyond single-tower VLMs to a hybrid reasoner-generator approach optimized for embodied AI and robotics. This is a direct challenge to closed-model dominance in physical reasoning and could reshape how enterprises build autonomous systems.
The key facts
5 to knowCosmos 3 architecture: two-tower mixture-of-transformers (autoregressive VLM + diffusion generator)
Capabilities: physical reasoning, world generation, action generation unified in single model
Released as open omnimodal world models
Targets physical AI / robotics use cases
Published June 3, 2026
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: NVIDIA released Cosmos 3, open omnimodal world models pairing an autoregressive VLM reasoner with a diffusion generator for physical AI. The post NVIDIA Releases Cosmos 3: A Two-Tower Mixture-of-Transformers Foundation Model Unifying Physical Reasoning, World Generation, and Action Generation…