· Industry· 日本語で読む
humanoid.guide publishes article covering Runway's 'Praxis 1'
humanoid.guide published an article covering Runway's 'Praxis 1.' The article does not report a new announcement about the model. Our previous articles have already covered its release terms and reported evaluation results.
What happened
humanoid.guide published an article titled 'Welcome, Praxis 1!' covering Runway's 'Praxis 1' on November 16, 2026.
Why it matters
Our previous articles covered the release terms for 'Praxis-1' and evaluation figures reported by Runway, so the latest article should be distinguished from a new performance demonstration. Separating information about a model's release from evaluations on physical robots is important for assessing its availability.
Analysis
What can be confirmed is that humanoid.guide published an article covering 'Praxis 1.' It should not be treated as a new product announcement by Runway or as an event in which the model's performance was independently verified. The article's publication and the model's technical progress therefore need to be considered separately.
We previously covered Runway's release of the 'Praxis-1' weights under Apache 2.0 on November 11, 2026, in Runway releases 'Praxis-1,' connecting video pretraining to robot control. The article's summary said Runway reported that the model, with 27B parameters, predicts video and actions simultaneously and achieved a 46.2% success rate in an evaluation on RoboLab-120. A report on November 12, Report: Runway releases 'Praxis-1,' a world action model for robots, also covered an article reporting the model's release. In light of this history, it is appropriate to place the latest article as another publication covering a model that has already been reported, rather than as the first report of its announcement.
The important point is not to equate released weights or benchmark figures with performance on individual robots. The summary of our November 11 article also focused on adaptation to each company's physical robots and the reproducibility of evaluations. Availability of an open model does not mean it will produce the same results on physical robots or maintain its performance in different environments. Model evaluations require distinguishing between figures reported by the announcing company, evaluation conditions and integration work on physical robots.
What remains to be confirmed is whether humanoid.guide's article presents new measurements or conditions, and whether it distinguishes independent evaluations on physical robots from figures announced by Runway. That cannot be verified from the article information provided. See humanoid.guide's original article for details.
Sources
Translated from the Japanese original: /news/humanoid-guide-runway-no-praxis-1-wo-atsukau-kiji-wo-keisai-afc106. The Japanese version is authoritative if they differ.