Physical Intelligence, the robotics lab behind the π0 and π0.5 robot foundation models, said on 13 October 2026 that it is making its latest models available for training and inference through a beta of the Pi API.
Developers upload their own robot data, launch fine-tuning on compute and infrastructure Physical Intelligence manages, and run the fine-tuned model on their robots through a hosted inference gateway, the company’s blog post says. Access is through a waitlist, and the post gives no pricing.
Business Pill · PHYSICAL AI
A one-minute explainer of physical AI: AI models that perceive and act in the physical world through robots. It teaches the general idea only and says nothing about any organisation in this story.
The key insight: As we read it, Physical Intelligence is moving from bespoke partnerships to a product: its post says developers can now run the collect, train, test and retrain loop themselves on its latest models, with the company running the compute and the hosted inference.
The post says adapting general-purpose robot models to a new task still requires ML infrastructure, compute and complex data pipelines, and that the Pi API is meant to handle those for the developer.
It says the latest model available through the API already provides more data-efficient fine-tuning and improved capability for its partners, and that the company plans to keep giving access to its latest and most capable models.
In its X post announcing the beta, Physical Intelligence says it has partnered with companies across diverse industries to adapt its models, from harvesting strawberries to assembling servers, and is now opening access to more teams.
Physical Intelligence’s stated Pi API workflow from its 13 October 2026 post: upload robot data, fine-tune a π model on compute it manages, and run the result through hosted inference. Access is by waitlist and the post gives no pricing.
What Was Posted on X
Announcing the beta release of the Pi API. We’ve partnered with companies across diverse industries to adapt our models to their robots and applications, from harvesting strawberries to assembling servers. Now we’re opening access to more teams.
— Physical Intelligence (@physical_int) Oct 13 21:08
Physical Intelligence announced the Pi API beta in this post on 13 October 2026. It says the company has partnered with companies across diverse industries to adapt its models, from harvesting strawberries to assembling servers, and is now opening access to more teams.
Summing up the challenge of commercializing robots in one photo
The bar for both reliability and output quality is much higher for physical labor than it is for knowledge work. A chatbot that produces work that is just “good enough” or low quality can still be useful.
A mediocre rough draft could still save time because a person can keep what’s good and fix the rest before its final. Copywriting, slides, and contracts can all be edited.
What is considered “good enough” for physical work has much higher standards. A bad weld can ruin metal. A wrong dose given to a patient is unacceptable.
Outputs from work done by humans or robots in the physical often serve as direct inputs into the workflow that are hard to recover from if an error arises. Outputs from work done by humans or Chatbots for knowledge work are much more tolerant of errors and poor quality because supervision can be baked in before the output is used further down the workflow.
The reliability threshold is 2-4 orders of magnitude greater for physical work than knowledge work and it’s what we’ve started to achieve this year. The upside is that by optimizing for reliability, robots will be able to work at a high level of utilization and operate at higher levels of autonomy compared to general purpose AI models in 2023. Machines that can deliver up to 24 hours of work a day as opposed to minutes or hours of work.
Investor Rewkang (Andrew Kang), posting on X on 27 October, argues the bar for reliability and output quality is much higher for physical labour than for knowledge work: a mediocre draft can be edited, but a bad weld can ruin metal and a wrong dose given to a patient is unacceptable. The post puts the reliability threshold for physical work at 2-4 orders of magnitude greater than for knowledge work, says that threshold has started to be achieved this year, and argues robots optimised for reliability could deliver up to 24 hours of work a day. These are the poster’s views.
Bloomberg: Physical Intelligence is in talks to raise at least $1B at roughly $20B, nearly double the $11.5B valuation from its earlier 2026 round.
Its valuation has climbed from $2.4B in November 2024 to $5.6B in November 2025, and the new round would roughly quadruple that figure in under a year.
The latest jump followed the October 13 Pi API beta, which gives outside developers hosted fine-tuning and inference on the company’s latest robot models through a waitlist.
Rohan Paul, summarising Bloomberg, wrote that Physical Intelligence is in talks to raise at least $1B at roughly $20B, nearly double the $11.5B valuation from its earlier 2026 round. He traced the valuation from $2.4B in November 2024 to $5.6B in November 2025, and noted that the latest jump followed the October 13 Pi API beta, which gives outside developers hosted fine-tuning and inference through a waitlist.
How the Loop Works
The post describes the same improvement process across its early collaborations: collect demonstrations, train a model, test it on the robot, and use the failures to decide what data to collect next.
The API is built to let developers run that iteration loop themselves, the post says: upload data, fine-tune a π model, run it on the robot through hosted inference, and use evaluation data to guide the next round.
It says most of the applications it shows, from Japan to Switzerland to deployments across the United States, ran with cloud-hosted inference through the API, and that some early deployments ran into network connectivity challenges.
Two Named Users
The post names Actor, which deploys autonomous heavy machinery, and says Actor’s autonomous wheel loader scoops material and loads it into a dump truck in a real commercial deployment with no operator inside the machine.
It also names Almond, which builds a dual-arm robot called Axol, and says Axol deburrs spacers used in its own assembly, running a model the team obtained after iterating with the Pi API for four days.
Physical AI has had a busy week in our coverage: Agility Robotics’ SPAC registration was declared effective on 13 October.
The Structural Read
As we read it, the API keeps the model and the infrastructure with Physical Intelligence while the customer brings the data: the post says developers upload data and access fine-tuned models through a hosted gateway.
As we read it, the feedback runs both ways: the post says the more its models are used, the more the company learns how to improve them.
As we read it, hosted inference brings its own constraint: the post says some early deployments ran into network connectivity challenges.
Physical Intelligence blog post, 13 October 2026
“What if the latest and most powerful robotic foundation models could be adapted to each application through a simple and ready-made API?”
Three Implications
ROBOT BUILDERS The post says the API lets them fine-tune Physical Intelligence models on their own data without building the ML infrastructure.
EARLY USERS The post names Actor and Almond, and says Almond got a working model after four days of iterating.
BUYERS The post gives no pricing, usage terms or general availability date; access is by waitlist.
The framework puts it this way: “The winning strategy in every major platform shift has followed a consistent pattern: Close the SCARCE layer (keep proprietary) + Open the ABUNDANT layer (commoditize)”.
As we read it, the Pi API sits on that line: the latest models, the fine-tuning compute and the inference gateway stay with Physical Intelligence, while access widens to more teams and their own robot data.
Physical Intelligence’s own account of two early users, from its 13 October 2026 post; we did not see the deployments.
What Is Not Established
The post gives no pricing, no usage terms and no general availability date, and it does not say how many developers are on the waitlist. It does not name which π model version sits behind the API.
The capability and fine-tuning claims are the company’s own; we did not use the API or see the deployments.
Update, 27 Oct: the funding round.Bloomberg reported that Physical Intelligence is in talks to raise at least $1 billion at a valuation of about $20 billion, with existing backers Avenir and Lux Capital in talks to lead the round. Physical Intelligence, Avenir and Lux Capital did not respond to Bloomberg’s requests for comment.
According to Bloomberg, the company secured funding this year that valued it at $11.5 billion, so the talks would put it at about 1.7 times that figure, by our arithmetic. Bloomberg reported that the company has no revenue.
Bloomberg tied the round to the beta described above: letting outside developers use and tailor its newest robotics models has increased investor interest in the fundraising, according to Bloomberg. It also named a rival marker: Skild AI reached a $14 billion valuation when it raised in January and said in September that its annual recurring revenue had passed $100 million.
The Bottom Line
Physical Intelligence says it has opened a beta of the Pi API, letting developers fine-tune its latest π robot models on their own data and run them through hosted inference. It names Actor and Almond as users and gives no pricing.
A note on sourcing. On 14 October 2026 we read Physical Intelligence’s blog post of 13 October in full, as text (we did not watch its videos), and its X post announcing the beta. Capability and user claims are the company’s own; we did not use the API or contact the company. Nothing here is investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
