A Guth Labs publication ↗
Back to News

Praxis-1 adds a low-cost robot-arm path, but consumer-GPU validation still matters

Praxis-1 adds a low-cost robot-arm path, but consumer-GPU validation still matters

Runway's Praxis-1 repository now lists SO-100/SO-101 checkpoints, including a quantization-aware-trained NVFP4 version. This materially extends the earlier DROID and bimanual release to another robot platform. The repository supplies a joint-space policy, server configuration and robot client. Primary repository

The extension was reported on November 13; the exact primary announcement timestamp was not independently recovered in this pass. This is a follow-up, not a fresh model-family launch or breaking news. Contemporaneous roundup

The hardware claim needs care. Runway's README still says peak memory on a 32GB RTX 5090 is unmeasured and that SM120 closed-loop testing has not occurred. It distinguishes the SO-101 checkpoint's quantization-aware training from simple post-training conversion. Availability should not become a claim of proven real-world reliability. Primary repository

Guth analysis: The useful advance is a concrete adaptation path for a more accessible arm. Before deploying it, developers still need platform-specific testing, calibration and safe operating boundaries. A smaller robot does not eliminate physical risk.

Evidence limits: Checkpoint pages and commit history could not be directly retrieved. Repository contents were read; weights were not downloaded or tested. No independent performance replication established. Earlier coverage is preserved rather than silently revised.

Sources and citations

Revision history

Published .