AI BriefingNVIDIA Expands Cosmos Platform with New Training Tools, Simulation Checkpoints, and Data Factory Blueprints
·2026.10.22 14:00
Key point
The expanded platform includes five cloud-native data blueprints, support for fine-tuning open models like GR00T and QWEN, and Cosmos 3 Nano variants that enable up to 2x faster inference.
Details
- NVIDIA expanded the Cosmos platform with new training tools, simulation checkpoints, and Cosmos Data Factory blueprints to help developers build world models for physical AI systems.
- Cosmos Data Factory now offers five cloud-native blueprints for autonomous vehicle data, image augmentation, and synthetic video generation, available via GitHub, Microsoft Azure, Nebius, and CoreWeave.
- Porsche is using Cosmos Data Factory through Voxel51 on Microsoft Azure to lead a shared AV data flywheel across Volkswagen Group brands, expected to reduce repeated physical test drives by up to 90% and shorten development cycles by up to 85%.
- Developers can adapt open models such as GR00T, Alpamayo, QWEN, and Pi0.5 through fine-tuning and reinforcement learning, with MicroAGI demonstrating a distillation of the 32-billion-parameter Cosmos 3 reasoner to an 8-billion-parameter Nano model with no loss in accuracy.
- New Cosmos 3 Nano model variants enable policy testing in multiview multisensor simulations, while updated Cosmos 3 NIM microservices deliver up to 2x faster inference with no quality loss in validated configurations.
- The new NVIDIA Metropolis Blueprint for video search and summarization (VSS) 3.3 helps teams monitor 40% more video streams with the same GPU, a capability ADNOC uses via Fogsphere’s MistIQ platform to detect safety hazards across 65 sites.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
RFM requires GPUs simultaneously across three stages: training, simulation, and edge inference