Cosmos at CES 2026 + turns 1 🎂

Big news from CES 🥳 + Cosmos turns 1 🎂

Cosmos Reason 2 is here — our most advanced reasoning vision-language model for physical AI, now topping the Physical AI Bench leaderboard 🏆

What’s new:

• Enhanced physical reasoning & spatio-temporal understanding
• Flexible deployment with 2B & 8B model sizes
• Long-context understanding (up to 256K tokens)
• Object detection with 2D/3D point localizations and trajectory data
• New Cosmos Cookbook recipes for faster onboarding

📖 Read the full blog | 👉 Download Cosmos Reason 2 | 🧑🏻‍🍳 Get Started with the Cosmos Cookbook

On top of Cosmos Reason 2 , we also rolled out other new updates, including:

• Cosmos Predict 2.5– A unified Text2World/Image2World/Video2World model trained on 200M clips, delivering higher-quality, better-aligned synthetic video worlds.
• Cosmos Transfer 2.5-2B – Lightweight, high-fidelity world-to-world translation with stronger physics alignment, fewer hallucinations, and more stable long videos.
• NVIDIA GR00T N1.6 – An open robot foundation model for general-purpose robotic learning and control, now available on Hugging Face and integrated with Cosmos Reason 1.