Physical Intelligence opens Pi API beta for Robot Foundation – Unite.AI models

Physical Intelligence announced the Pi API beta on October 13, 2026, making its latest robotics base models available to external developers for hosted tuning and inference. Access to the beta is through a waiting list, where applicants are asked to describe what they want to build.

According to the workflow established in the announcement, a developer uploads the robot’s data, starts tuning the computation and infrastructure managed by Physical Intelligence, and accesses the optimized model through a hosted gateway to run inference on their robot. The evaluation data then guides the next iteration. The company said that the latest model available through the API already offers more efficient data tuning and better capacity for its partners, and that because it can quickly update the models underlying the API, it plans to continue enabling access to its newer, more capable models. Physical Intelligence described the launch as the first step towards providing a physical intelligence foundation for the community to build on and enabling end-to-end autonomy.

The models behind the API

The announcement defines generic robotic baseline models like π0 and π0.5 as a promising way to address the challenge of deploying robots: in addition to identifying the right use case and building the right hardware, developers need models that are intelligent enough to perform the task at hand. These models provide a solid foundation of knowledge and physical capabilities that substantially reduce the data and effort needed to adapt to new tasks, environments or robots, the company said, but adapting them still requires complex ML infrastructure, processing and data pipelines.

Physical Intelligence publishes open source versions of these models in its openpi repository. The repository contains three types of models: π0, a flow-based vision-language-action model; π0-FAST, an autoregressive vision-language-action model based on the FAST action tokenizer; and π0.5, an updated version of π0 with better open-world generalization, trained with a method the company calls knowledge isolation. The baseline checkpoints for all three are pre-trained on over 10,000 hours of robot data, and the repository is released under the Apache-2.0 license. Physical Intelligence released π0.5 and PyTorch support for the repository in September 2025, after adding instructions for training vision-language-action models on the entire DROID dataset in June 2025.

Running open models locally requires an NVIDIA GPU with more than 8GB of memory for inference, with the RTX 4090 provided as an example, and more than 70GB of memory on hardware like the A100 or H100 for full tuning. The repository was tested on Ubuntu 22.04 and does not currently support other operating systems. It also supports remote inference, where a model runs on a separate server and transmits actions to the robot via a websocket connection. The repository notes that π0 was developed for the company’s robots, which differ from widely used platforms such as ALOHA and DROID, and that the team does not expect every adaptation attempt to be successful. The API takes that pipeline as a hosted service: users provide the data, and Physical Intelligence handles the training computation and serves the finished model through its gateway.

Partner implementations across all industries

Physical Intelligence said it sought to answer two questions with the API: whether its models support a wide range of applications and robotic hardware across different industries, and how easily the API allows developers to iterate toward useful and practical autonomy. To test breadth, the company works closely with companies deploying robots, using these collaborations to validate how its models perform across different tasks, hardware and operating environments.

The announcement concerns Actor, a Utah-based company that distributes autonomous heavy machinery. In the footage described by the company as a real commercial use, Actor’s autonomous wheel loader collects material and loads it onto a dump truck without any operator inside the machine. Other collaborations mentioned in the announcement include AgRobotics, CloudChef, a dairy partner, Ultra and Weave.

In these initial collaborations, the company said, a common pattern emerged. The challenges varied from task to task, from a spilled garment to an ingredient that wouldn’t stay in the scoop to a production part that ended up in the wrong place, but the improvement process was more or less the same: collect demonstrations, train a model, test it on the robot, and use the errors to decide what data to collect next. Physical Intelligence said it created the API to allow developers to perform this iteration loop themselves, using its latest models.

Early Access results and next steps

To test the independent iteration, Physical Intelligence gave a small number of companies early access to the API and said it was impressed with how quickly they got started and began making progress. The announcement highlights San Francisco, California-based Almond, which builds Axol, a dual-armed robot for physical AI builders whose stated goal is for Axol to build itself. According to the company, Axol deburred the spacers used in its assembly while running a model obtained by the team after iterating the Pi API for four days.

Other early access users shown include Argon Robotics, Chef Robotics, Deft, Dream Machines, Formic, Foundry, Frontier Machines, General Instinct, Intuition.dev, Omakase, robot.com, Ultra, and Yondu, along with a hackathon project. Most of the applications shown ran with inference hosted in the cloud via the API, spanning deployments from Japan to Switzerland to sites in the United States. The company said these early deployments covered a range of environments and tasks and highlighted areas for improvement, including network connectivity challenges for some.

Physical Intelligence said the examples show that the models it develops can be used for real-world, useful applications, and that its partners are able to use the API to independently iterate policies and improve autonomy in their tasks, while characterizing the effort as still early, with developer feedback set to help shape future releases. The company said it is hiring researchers, infrastructure, hardware, people and operations.