Runway has released Praxis-1, a 27-billion-parameter open-weight model designed for predicting future video and robot actions. This model achieved a 46.2% success rate on NVIDIA's RoboLab-120 leaderboard, outperforming other models like FLUX 3 Action and Cosmos3 Nano. The model's architecture combines a diffusion-transformer video backbone with an action expert, and it can generate 1.6 seconds of motion in under a second on an NVIDIA H100 GPU. The article provides a checklist for developers considering using Praxis-1, emphasizing hardware sizing, cost budgeting, and the importance of fine-tuning and rigorous testing before deploying on physical robots. AI
IMPACT Sets new SOTA on robot action benchmarks; may accelerate adoption of open-weight models for robotics.
RANK_REASON Frontier-lab model release with system card [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- Cosmos3 Nano
- DROID
- FLUX 3 Action
- Hugging Face
- Jetson Thor
- LeRobot
- NVIDIA
- NVIDIA H100
- Praxis-1
- RoboLab-120
- RTX 5090
- Runway ML
- Shivam Kumar
- vllm-omni
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →