Runway has released Praxis-1, its robot controlmodel, with open weights under the Apache 2.0 license, the company announced on 11 November. Runway says the model reaches 46.2% success on NVIDIA’s RoboLab-120 leaderboard against 28.0% for π0.5, about 1.65 times as high by our arithmetic. The scores are Runway’s own figures.
Runway calls Praxis-1 a world action model: a 27B-parameter mixture-of-experts diffusion transformer “that predicts video and actions together”. The license covers the weights and the inference and fine-tuning code, “including for commercial use”, according to the announcement.
Business Pill · Physical AI Explained
An evergreen Business Engineer explainer on physical AI. It teaches the general idea only and does not discuss Praxis-1.
The key insight: Runway has released its Praxis-1 world action model with open weights and code under Apache 2.0, including for commercial use, and reports 46.2% success on NVIDIA’s RoboLab-120 leaderboard against 28.0% for π0.5, about 1.65 times as high by our arithmetic. The scores are Runway’s own.
Runway introduced Praxis-1 in September and said then that it would ship open weights. The release puts the model on GitHub and on Hugging Face with LeRobot support, in vLLM-Omni, and on Reactor for hosted inference. Hugging Face lists eight Praxis-1 repositories under Runway’s account, including base, DROID and YAM checkpoints.
Runway says Praxis-1 runs on NVIDIA hardware from H100 and B200 data center GPUs to RTX 5090 and RTX PRO 6000 workstation cards and Jetson Thor on the robot. Launch partners Noble Machines, Standard Bots and Ultra have run it on a legged humanoid, a six-axis industrial arm and a dual-arm warehouse robot, the company says.
Success rate across 1,200 episodes on NVIDIA’s RoboLab-120 leaderboard, as reported by Runway on 11 November 2026: Praxis-1 46.2%, FLUX 3 Action 42.9%, Cosmos3 Nano 36.8% and π0.5 28.0%. The figures are company-reported; the 1.65x ratio and point gaps are our arithmetic.
The Scores, as Runway Reports Them
On RoboLab-120, Runway reports Praxis-1 at 46.2% success across 1,200 episodes, against 42.9% for FLUX 3 Action, 36.8% for Cosmos3 Nano and 28.0% for π0.5. By our arithmetic, that is 3.3 points ahead of FLUX 3 Action and 18.2 points ahead of π0.5. Runway calls it “the top-performing world action model on NVIDIA’s RoboLab-120 leaderboard”.
Fine-tuned on Google’s RT-1 robot, which Runway says never appeared in its training data, Praxis-1 reaches 80.0% success on SimplerEnv against 52.4% for the original RT-1 policy, a gap of 27.6 points by our arithmetic. Our piece on PokeBot’s free robot API covered another company’s self-reported comparison with π0.5.
Why Runway Says It Opened the Weights
Runway writes that “U.S. leadership in physical AI will depend on investment in both hardware and software”, and that open weights give hardware developers the flexibility to run, inspect and adapt a world action model on their own terms. It describes itself as a founding member of the Cosmos Coalition with NVIDIA.
The company also says its research shows “web video matches teleoperated robot video as pretraining data for policy performance”. Our piece on how NVIDIA’s GR00T 2 trains 92% on human video covers a related approach.
The Structural Read
As we read the announcement, Runway is releasing the model itself free to use under Apache 2.0, while pointing developers to NVIDIA hardware from data center GPUs to Jetson Thor on the robot and to hosted inference on Reactor. Its stated reason is that open weights let hardware developers run, inspect and adapt the model on their own terms.
Runway, announcing the open-weight release of Praxis-1, 11 November 2026
“Open weights give hardware developers the flexibility to run, inspect and adapt a world action model on their own terms”
Three Implications
FOR ROBOT MAKERS The license covers the weights and the inference and fine-tuning code, with support for both full fine-tuning and LoRA, according to the announcement.
FOR THE BENCHMARK RACE Runway reports Praxis-1 3.3 points ahead of FLUX 3 Action and 18.2 points ahead of π0.5 on RoboLab-120, by our arithmetic.
FOR NEW HARDWARE Fine-tuned on Google’s RT-1 robot, Praxis-1 reaches 80.0% on SimplerEnv against 52.4% for the original RT-1 policy, Runway says, a 27.6-point gap by our arithmetic.
The framework says: “Close the SCARCE layer (keep proprietary) + Open the ABUNDANT layer (commoditize)”.
As we read it, Runway has opened the model layer of robot control while the announcement points to hosted inference on Reactor and to NVIDIA hardware for running it. The framework predates the release and does not discuss it.
SimplerEnv success on Google’s RT-1 robot: the original RT-1 policy against Praxis-1 after fine-tuning, as reported by Runway. The 27.6-point gap is our arithmetic.
What Is Not Established
The RoboLab-120 and SimplerEnv figures are Runway’s own and are self-reported; we have not seen an independent replication. The announcement compares Praxis-1 with three named models; we have not seen the full leaderboard. A success rate in a benchmark suite is not a measure of performance on a customer’s robot or tasks.
Runway was not contacted for this piece.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Runway has released Praxis-1 with open weights and code under Apache 2.0, including for commercial use. It reports 46.2% success on NVIDIA’s RoboLab-120 leaderboard, about 1.65 times π0.5’s 28.0% by our arithmetic, and 80.0% on SimplerEnv after fine-tuning on RT-1, against 52.4% for RT-1’s original policy.
A note on sourcing. Model details, license terms, partners and benchmark scores come from Runway’s announcement of 11 November 2026, which we read in full, and its September research page; the repository count is from Hugging Face’s listing for Runway’s account, read on 11 November. The RoboLab-120 and SimplerEnv scores are Runway’s own figures. Ratios and point gaps are our arithmetic. Runway was not contacted for this piece.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
