Physical Intelligence opens a waitlisted API beta so other companies can fine-tune and run its robot models

Physical Intelligence, the San Francisco lab behind the π family of robot foundation models, published “Frontier Models for Every Robot” on 13 October. Its first sentence: “We are making our latest models available for training and inference through the Pi API beta.” Customers supply data from their own robots, Pi fine-tunes one of its models on infrastructure it manages, and the resulting policy runs through its API.

Access is by waitlist: “If you would like to participate in the beta, please tell us what you want to build.” The post says a small number of companies had early access. It describes two in its text. Actor's “autonomous wheel loader scoops material and loads it into a dump truck during a real commercial deployment without any operator inside the machine”; the video caption gives the location as Utah. The text also names Almond; other early users appear only as logos.

Pi says most of the applications shown, from Japan to Switzerland and across the United States, ran on cloud inference through the API. It gives no prices, success rates or list of supported robots, and it is candid about the stage: “We're still early, and feedback from developers will help shape future releases.”

The change is in the business model. Pi had published research and open weights and worked with a few named partners, such as Ultra, whose funding the Switchboard ran on 10 October. Now it is selling hosted training and inference to any robot maker. The same day, Google pitched Gemini 4 Argon to robot builders.

• Claimed Pi says it is making its latest models available for training and inference through the Pi API beta, with access by waitlist. Physical Intelligence
• Claimed Pi says Actor's autonomous wheel loader loads a dump truck during a real commercial deployment with no operator inside. Physical Intelligence
• Confirmed The post gives no pricing and says the product is still early. Physical Intelligence