Physical Intelligence develops general-purpose AI models for robots, called vision-language-action models, that turn camera images and instructions into movements. Its first model, π0, was followed by π0.5, π*0.6 and π0.7, and the company publishes code and weights for its earlier models in the openpi repository. Its latest models are offered through the Pi API beta, where developers upload data, fine-tune and run hosted inference on their own robots. The company is based in San Francisco.
Main products
- π0 and π0-FAST
- π0.5
- π*0.6
- π0.7
- Pi API (beta)
- openpi
Pricing model
The Pi API beta, announced on 13 October 2026, is open through a waitlist and its announcement states no prices. The openpi repository on GitHub (Apache 2.0) publishes code and base checkpoints for π0, π0-FAST and π0.5. Official pricing page.
Known limits
- Offers models and an API, not robots: customers bring their own hardware
- Most applications shown in the API announcement ran on cloud-hosted inference, and Pi reports network connectivity problems in some early deployments
- API access is a beta behind a waitlist
Alternatives in the directory
- Figure AI: Developer of the Figure 03 general-purpose humanoid robot and Helix, the AI model that runs on board
- NVIDIA: Jetson system-on-modules and software for robotics, vision and autonomous machines
- AgiBot: Shanghai developer of A, X and G series humanoid robots, D1 quadrupeds and open embodied AI datasets
Products, pricing and limits checked against the company's official site on October 23, 2026. Prices and plans change; the official page is the reference. Spotted an error? Tell us or see the editorial policy.
Focus: vision-language-action models, robot foundation models.
Tags: embodied ai · vla · robotics · foundation models · open source · physical intelligence
More edge ai companies
Alif Semiconductor
Ensemble and Balletto microcontrollers with Arm Ethos NPUs for battery-powered AI
Ambarella
Computer vision SoCs with on-chip neural network engines for cameras, robots and vehicles
Aspinity
Analog machine learning chips that classify sensor data before it is digitized