Being-H0.7
$ 0
Being-H0.7 is BeingBeyond’s world-action model that learns from huge amounts of human video. It plans ahead in a hidden space and drives dexterous robot hands in real time.
Available on backorder
Brain Score
5
Specifications and details:
| Nationality | China |
|---|---|
| Website | https://beingbeyond.com/en/ |
| Model type | Foundation Model |
| Manufacturer | BeingBeyond |
| Release date | April 14, 2026 |
Description
Being-H0.7 is a world-action model from BeingBeyond, a Beijing startup that builds robot foundation models. The team released it in April 2026 as the follow-up to Being-H0.5. It learns from about 200,000 hours of first-person human video and 15,000 hours of robot demonstrations. During training, it learns to picture what comes next in a compact hidden space. However, it never draws actual future video frames at run time. So it keeps the planning benefit of a world model while staying fast enough for live control.
The team tested it on three robots: the PND Adam-U, the Unitree G1 and a Franka arm, all fitted with Linkerbot five-fingered hands. For example, the robots moved liquid with a pipette, folded clothes and sorted items into drawers over several steps. On the G1, a separate leg controller keeps the humanoid balanced while the model drives the arms. Meanwhile, each control step takes only a few milliseconds, and the robot keeps moving while the next plan loads. BeingBeyond published the paper and project page, and it says code and checkpoints are coming soon. NeurIPS 2026 also accepted the paper for its main program.
New! 2026 Humanoid
Robot Market Report
198 pages of exclusive insight from global robotics experts — uncover funding trends, technology challenges, leading manufacturers, supply chain shifts, and surveys and forecasts on future humanoid applications.
now Google DeepMind
Contact Humanoid.guide
