Perceptron released Perceptron Egocentric 1.0 on 4 November, “a self-service API that turns first-person human and robot video into structured training data for robotics”, its launch blog says. Perceptron lists subtask timelines at $9 per hour of video and estimates that OpenAI’s GPT-6 Astra costs about $76 for the same timelines: about 12%, or roughly one-eighth, by our arithmetic.
On EgoHome, a benchmark Perceptron built itself, the company reports a score of 0.787 at its default setting against 0.797 for GPT-6 Astra at 4 frames per second, a gap of 0.010, by our arithmetic. Perceptron’s X post at 16:34 UTC summed it up as “Fine-grained subtasks matching Astra quality at ~1/8 the cost”.
Business Pill · Physical AI Explained
A short Business Engineer explainer on physical AI, models that perceive and act in the real world. It teaches the general idea only and does not discuss Perceptron.
The key insight: Perceptron lists subtask timelines for first-person robot-training video at $9 per video-hour and estimates GPT-6 Astra’s cost for the same output at about $76, roughly one-eighth by our arithmetic, while its own EgoHome benchmark puts the two within 0.010.
The API segments first-person video into timestamped subtasks, labels what each hand or robot gripper is doing over time, and tracks hands in 2D or 3D. Customers can “Send up to 1,000 video links in one request”, then poll each job and download its results. Perceptron calls it “a major upgrade from our early-access launch in July”.
List prices are per hour of video: $9 for the subtask timeline, which every job includes, plus $12 for 2D hand tracking, $14 for 3D hand tracking and $6 for per-hand activity labels. With 3D tracking and activity labels, that adds up to the $29 per video-hour the blog cites, by our arithmetic.
Circles sized by area: cost to produce subtask timelines for one hour of video. Perceptron’s $9 is its list price; the ~$76 for GPT-6 Astra is Perceptron’s own estimate. EgoHome scores are from Perceptron’s own benchmark. Source: Perceptron blog, 4 November 2026; not independently verified.
What Was Posted on X
Today we’re releasing Perceptron Egocentric 1.0: The frontier for egocentric annotations, now through a self-service API. • Fine-grained subtasks matching Astra quality at ~1/8 the cost • State-of-the-Art 2D and 3D hand tracking • Individual timelines for every action on each hand and robot gripper
Perceptron’s launch post, 16:34 UTC on 4 November. It announces Perceptron Egocentric 1.0 as a self-service API and claims fine-grained subtasks matching Astra quality at about one-eighth the cost, plus 2D and 3D hand tracking and per-hand and per-gripper timelines.
How the $76 Comparison Is Built
The Astra figure is Perceptron’s estimate, not an OpenAIprice. The blog says it counted “only the timeline’s share: one pass of the video as input plus the timeline output”. At 8 frames per second, Perceptron scores 0.798, level with Astra’s 0.797, at an estimated $15.
Perceptron also says its 3D hand model reconstructs hand shape “12–25% more accurately than ACE-Ego-Hand”, the strongest prior method in its table, and needs no camera calibration. On steps shorter than half a second, it says GPT-6 Astra at 1 frame per second and NVIDIA’s FLORA score slightly higher than its own model.
Connecting the Coverage
GenRobot, which turns recordings of people doing tasks into robot training data, raised $100 million, TechNode reported, citing Sina Finance; PEdaily reported that its deliveries rose 10x to 1 million hours in a single quarter. At Perceptron’s $9 list price, timelines for that many hours would cost $9 million before volume discounts, by our illustrative arithmetic; the sources do not say GenRobot uses Perceptron.
As we read the launch, the product is priced per hour of video, so the comparison that matters to a buyer is cost per labelled hour, and Perceptron puts its own on the same table as a frontier model.
As we read the caveats, the claim rests on Perceptron’s own benchmark and its own estimate of Astra’s share of a bill; EgoHome has not been released yet, so outsiders cannot yet rerun it.
Perceptron, Introducing Perceptron Egocentric 1.0
“Frontier models like GPT-6 Astra produce strong subtask annotations but cost too much to run on large datasets”
Three Implications
FOR ROBOTICS TEAMS Perceptron lists $9 per video-hour for subtask timelines, with 2D or 3D hand tracking and per-hand activity labels as paid add-ons, through a self-service API.
FOR DATA BUYERS At list price, 1,000 video-hours of timelines would cost $9,000, against about $76,000 at Perceptron’s Astra estimate, by our illustrative arithmetic.
FOR READERS OF THE NUMBERS The scores come from Perceptron’s own runs and its own benchmark, and the Astra cost is its estimate, not an OpenAIprice.
The framework says: “Each moat is built by a different data pipeline.”
As we read it, Perceptron is selling one step of a robotics data pipeline, turning first-person video into labelled subtasks and hand tracks, at a list price per hour of video. The framework does not discuss Perceptron. For the full framework, read The Four Intelligence Moats on businessengineer.ai.
Every score here is Perceptron’s own. EgoHome is its own benchmark and has not been released yet; the blog says it will follow “in the coming days”. The Astra cost is Perceptron’s estimate of one task’s share of a bill. Perceptron did not name customers in what we read.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Perceptron is selling robot-training video labels through a self-service API at $9 per video-hour for subtask timelines. By its own benchmark and estimate, that comes within 0.010 of GPT-6 Astra’s score at about one-eighth of Astra’s roughly $76 cost.
A note on sourcing. Product details, prices, benchmark scores and quotations are from Perceptron’s launch blog and its X post of 4 November 2026. The GPT-6 Astra cost is Perceptron’s estimate, and EgoHome is Perceptron’s own benchmark; we have not reproduced any result. Ratios and sums are our arithmetic.
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.
