Runway unveils Praxis-1, AI engine outperforming gpt-6 astra in robotics tasks by significant margins

Runway, a US video AI company, introduced its open-weight robotic model, Praxis-1, on November 11, 2026. This model leverages knowledge learned from video pre-training to control real-world robots, addressing the high cost associated with collecting robot training data. The model is available under the Apache 2.0 license, allowing developers to run and fine-tune it on their own hardware. Built upon Runway's 27-billion-parameter video foundation model, Praxis-1 supports local deployment, including an NVFP4 quantized version optimized for the NVIDIA RTX 5090. Testing on the RoboLab-120 benchmark showed the model achieved a success rate of approximately 46.2% across 555 out of 1200 scenarios, leading in the world action and open robot foundation model categories. Runway stated that Praxis-1 demonstrates substantial advantages over current state-of-the-art robotic arms. In identical workstation and task settings, Praxis-1 operates 15.6 times faster than a robot driven by GPT-6 Astra, and its operational cost is 51 times lower. The company noted that research is still in its early stages, with future efforts focusing on long-term planning and self-correction based on predictive outcomes.