Runway open-sources Praxis-1, claiming 1.65x π0.5 on Nvidia's test

Runway has open-sourced Praxis-1, a model that reaches 1.65x the performance of π0.5 on Nvidia's test, according to FourWeekMBA.

π0.5 is one of the more widely cited baselines in robot foundation models, so using it as the reference point places Praxis-1 in the embodied AI conversation rather than in plain text or image generation.

The report hangs on two words: open source, which decides whether the community can reproduce the result, and 1.65x, which decides whether it is worth reproducing.

For models like this, open source means more than downloadable weights; it is about whether training and evaluation details survive scrutiny from outside. In robotics, differences in task suites, action spaces and hardware configuration often matter more than the weights themselves when judging a comparison.

It is worth noting that the information comes from a headline-level report, and the test conditions, task suite and baseline configuration are not spelled out in the available material. The 1.65x figure is therefore better read as a public claim awaiting verification than as a completed head-to-head evaluation.

A new open baseline lowers the bar for comparison in robotics: newcomers no longer start from zero, they start against a number already on paper and have to explain the gap.

What to watch is whether the community can reproduce the 1.65x figure on its own hardware, and whether Runway publishes the full evaluation protocol and training details.