Generalist raises $198M for robot models at a reported $3B valuation

Generalist AI Inc. has sold $198.2 million of stock, according to a Form D the robotics startup filed with the Securities and Exchange Commission on August 24. The filing puts the full offering at $208.2 million, leaves just under $10 million unsold, and records 32 investors who’d already taken part.

The filing sets the amount, and the reporting sets the price. TechCrunch reported on August 25 that the raise values Generalist at $3 billion and that 8VC led it, citing two people with knowledge of the funding. Neither figure appears in the Form D, which names no investors and states no valuation.

That matters because this is an extension rather than a fresh round. TechCrunch says the money extends the $400 million Series B that Radical Ventures led at a $2 billion valuation in June, taking the round’s total to $600 million. Generalist’s own June announcement put that raise at $400 million and total funding at “more than half a billion dollars”, naming Radical Ventures as lead alongside 8VC, Union Square Ventures, Hanabi Capital and Norwest. By our arithmetic the paper price is up 50% in under three months, which is exactly the sort of detail a round announcement tends to leave out.

Stage	When	Amount	Valuation
Series B, led by Radical Ventures	Announced 4 June 2026	$400M	$2B, per TechCrunch
Extension, reportedly led by 8VC	First sale 7 August, filed 24 August	$198.2M sold of a $208.2M offering	$3B, per TechCrunch
Round total		$600M, per TechCrunch

The dates are worth reading slowly, because they don’t quite line up. SiliconANGLE reported that the raise came about a week after Generalist debuted its newest model, and that Axios was first to name 8VC as the lead. The filing tells a slightly different story: it dates the first sale to 7 August, twelve days before the company published the model on 19 August. So the money was moving before the results were public.

The model is GEN-1.5, and the claim behind it is speed of teaching rather than raw skill. Generalist’s write-up says the model learns a new task from 3 to 12 seconds of a single demonstration dropped into its context window, with no fine-tuning and no gradient updates. It takes video with 30 seconds of memory, plus sensor and proprioceptive input, and outputs action trajectories at 100 Hz.

Across 10 tasks the company reports 59% average success from that single example, with a standard deviation of 10 points. Ten gradient steps on five minutes of data per task, roughly 50 demonstrations, lift the average to 83%. Generalist also says GEN-1.5 has been pretraining continuously for over eight months. SiliconANGLE argues a sizable portion of the new capital goes towards AI infrastructure for that reason, while noting the report it cited didn’t say how the money gets spent.

Although the tasks are simple and short-horizon, and the success rates are modest, this is the first model we know of that has demonstrated the general ability to learn a wide range of dexterous closed-loop physical tasks from just one-shot or few-shot demonstrations.

Generalist Team, GEN-1.5 research post, 19 August 2026

Read that caveat as the company wrote it, because it does a lot of work. These are Generalist’s own numbers, on tasks Generalist chose, with no outside audit attached, and the firm calls them simple and short-horizon itself. The distance between a bench with a robotic arm on it and an unfamiliar room is the part nobody has closed.

Not everyone in robotics thinks that distance closes soon. At the 2026 World Robot Conference in Beijing, Unitree chief executive Wang Xingxing said the field reaches its “ChatGPT moment” only when a robot put into a home it has never seen can finish about 80% of tasks from spoken commands. Business Insider reported he put that at two or three years, or five to 10 if progress slows.

Even so, $3 billion is a modest number in this cohort. TechCrunch puts Physical Intelligence at a reported $11 billion and SoftBank-backed Skild AI at $14 billion, both chasing the same idea of one brain across many robot bodies. Generalist was founded in 2024 by former Google DeepMind researchers Pete Florence and Andy Zeng with former Boston Dynamics engineer Andrew Barry, and it stayed quiet until recently.

So two things are worth watching from here. The filing still shows $9,999,908 of the offering unsold, so the final figure isn’t fixed yet, and Generalist hasn’t published anything about this raise on its own blog. Until it does, the valuation rests on two anonymous sources and the amount rests on a filing that says nothing about price.