hebel. today · 13 November 20263 / 80

Technology · AI & Robotics

Robotics models shift toward open weights and faster on-robot inference

Runway released Praxis-1, a 27-billion-parameter open-weight world action model that ranks first on the RoboLab-120 leaderboard with a 46.2% success rate. By predicting video and robot actions simultaneously, the model enables robots to generate…

Runway released Praxis-1, a 27-billion-parameter open-weight world action model that ranks first on the RoboLab-120 leaderboard with a 46.2% success rate (TLDR AI). By predicting video and robot actions simultaneously, the model enables robots to generate movement at 15 hertz from 24 action steps per call (Sifu Yik's Substack). The model operates on hardware ranging from data-center GPUs to the Jetson Thor chip inside a robot, completing tasks 15.6 times faster than GPT-6 Astra at a 51 times lower cost (Latent Space).

  • Google DeepMind’s Gemini Robotics 2.5 improves task success rates by up to 100% and completion speeds by 30% across 200 tested activities (AI Weekly).
  • NVIDIA’s GR00T 2 Turbo model predicts actions in 124 milliseconds on the Jetson AGX Thor, allowing for real-time control on mobile robots (Michelle Sun from Core Matter).
  • Walden Robotics reports its deployed humanoids have successfully loaded 33,000 camshafts on a Toyota production line since April (Michelle Sun from Core Matter).
  • Robbyant’s open-source LingBot-Touch model integrates tactile data from 5,000 hours of recordings to help robots sense force and grip security (The Page Between).
  • Tapestry’s HyperQ agent cleared site-control paperwork for 220 gigawatts across 811 applications in under a day, compared to two months of manual work (Ben Linford from Shared Sapience).

Try today · 45 min

Test the Praxis-1 model on your local hardware using the provided LeRobot integration.

Download the base model weights from the Runway GitHub repository to begin inference.

You need

  • NVIDIA GPU (H100 or Jetson Thor)
  • LeRobot library
  • Praxis-1 model weights

Steps

  1. Clone the Runway Praxis-1 repository from GitHub.
  2. Install the LeRobot dependencies on your machine.
  3. Load the base model checkpoint.
  4. Run the provided inference script on a sample task.

One email instead of your newsletter pile.

Forward the newsletters you already get (a Gmail filter does it) and tell Hebel your goals. Every morning it reads them with 20,000 others and sends one short email, with one small thing to do today. Fewer emails, not more. Free during beta.

Start my paper →
LinkedIn X

Sources

  1. Latent Space on hebel. →This digest details the performance metrics of Praxis-1 compared to GPT-6 Astra on robotic arm control.

  2. Michelle Sun, Core Matter on hebel. → ✎ Written by a personThis recap covers the efficacy of deployment data and the operational results of Walden Robotics humanoids.

  3. AI Weekly on hebel. →This report highlights the 96% success rate achieved by Gemini Robotics 2.5 in automotive part-sorting trials.

  4. Sifu Yik's Substack on hebel. →This article provides the technical specifications and download instructions for the open-weight Praxis-1 model.

  5. The Page Between on hebel. →This newsletter describes the LingBot-Touch model and its role in translating tactile sensor signals for robot actions.

  6. TLDR AI on hebel. →This summary lists the performance of Praxis-1 on the RoboLab-120 benchmark and its compatibility with NVIDIA hardware.

  7. Ben Linford, Shared Sapience on hebel. → ✎ Written by a personThis report documents the use of AI agents to automate regulatory paperwork, reducing processing time from months to one day.

More in Technology

Autonomous vulnerability discovery ends the two-week patch cycle

Consolidating daily operations into a single text-based dashboard

Developers build local-only dashboards using multi-agent teams

Europe's startup gap: Structural decay or AI-driven bypass?

API gateways consolidate cross-cutting concerns to simplify microservice communication

Data brokers face state-level deletion mandates

AI & Robotics yesterday

Gemini Robotics 2.5 hits 96% success in Mercedes-Benz pilot

Sign in to Hebel first

Likes, your collection and done Hebel need an account. We email you a six digit code; no password.