Generalist

Generalist (also referred to in press as Generalist AI) is an AI research and product company building what it calls general intelligence for the physical world. Rather than manufacturing robots, it develops embodied foundation models — the "brains" — designed to run across different robot bodies and environments, from factories and laboratories to homes and space. Its stated focus is dexterity, which it says requires simultaneous advances in data, models, and hardware.

The company's public model line began with a June 2025 research preview of end-to-end neural networks for dexterous sensorimotor policies, followed by GEN-0 in November 2025, GEN-1 in April 2026, and GEN-1.5 in August 2026. GEN-0 is described as trained directly on high-fidelity raw physical interaction rather than primarily on vision-language pretraining, and introduces "Harmonic Reasoning," an approach in which asynchronous continuous-time streams of sensing and acting tokens let the model think and act simultaneously without System1-System2 architectures or inference-time guidance. The company reports scaling laws in which pretraining data and compute predictably improve downstream post-training performance, and a phase transition around 7B parameters, below which models exhibit ossification under data overload; GEN-0 was scaled past 10B parameters and tested on 6DoF, 7DoF, and 16+DoF semi-humanoid robots. GEN-1 is stated to have roughly 99% of its parameters trained from scratch and is positioned by the company as neither a vision-language-action model nor a world model, but a native foundation model for physical interaction.

Publicly demonstrated capabilities include folding laundry, sorting Lego bricks, whisking baking ingredients, one-shot Lego assembly from an example structure, and a long-horizon camera-kit packing task. The company ran a live demo at GTC in March 2026.

Founding story

Generalist was founded in 2024 by former Google DeepMind robotics researchers Pete Florence and Andy Zeng together with former Boston Dynamics roboticist Andrew Barry. Florence, the CEO, was previously a senior scientist at DeepMind where his work included RT-2, an early vision-language robotics model; Barry serves as CTO and came from Boston Dynamics.

Business model

Generalist develops and sells robot intelligence rather than robots: cross-embodiment foundation models intended to be deployed on third-party hardware across industrial and domestic settings. It also operates its own large-scale data collection apparatus, including hand-mimicking gripper devices distributed to contributors globally, as an input to model training.

Not disclosed in the sources. The company has stated it has customers using GEN-1 but has not identified them, and describes GEN-1 as beginning to cross into commercial viability for very simple, short (typically sub-minute) tasks.

Traction

Public traction indicators include an in-house dataset that grew from over 270,000 hours (November 2025) to over 500,000 hours (mid-2026) of real-world interaction data, a cadence of model releases from GEN-0 through GEN-1.5 within roughly ten months, unnamed customers using GEN-1, more than half a billion dollars raised in total, and a valuation that reportedly moved from $2 billion in June 2026 to a discussed $3 billion by late July 2026.

Latest developments

In August 2026 the company published GEN-1.5, presented as an embodied foundation model that can learn new tasks in context from about 12 seconds of demonstration. In July 2026 it described GEN-1 support for a broad range of end effectors, and was reported to be negotiating a new round at a $3 billion valuation led by 8VC, one month after announcing a $400 million round at a $2 billion valuation.