Generalist Secures $200 Million in Funding with Support from Jensen Huang, Fei-Fei Li, and Bin Lin

Wow, the Generalist funding progress bar must have a 2x speed cheat—this growth is incredible!

I’m not exaggerating—according to Axios, less than three months after Generalist raised $400 million (RMB 2.87 billion) in its Series B round and became a unicorn, it quietly completed another funding round of $200 million (approximately RMB 1.44 billion), pushing its latest valuation above $2 billion (approximately RMB 14.36 billion).

NVIDIA, Bezos Expeditions led by Jeff Bezos, and Fei-Fei Li, known as the "Godmother of AI," are all listed among its shareholders.

Generalist is an AI company founded in 2024 in the San Francisco Bay Area, USA, focused on building a general-purpose brain for robots. It was co-founded by former Google DeepMind researchers Pete Florence and Andy Zeng, and former Boston Dynamics engineer Andrew Barry.

△

According to Forge, as of the time of writing, Generalist has completed four rounds of funding, raising approximately $750 million (about RMB 5.39 billion), with investors including NVIDIA, Boldstart Ventures, Lin Bin, co-founder of Xiaomi, Li Fei-Fei, Naval Ravikant, and Yuan Zheng, founder of Zoom.

Generalist neither showcases flashy humanoid robots nor relies on robotic arm orders to tell stories. Instead, they have quickly risen to prominence by leveraging their team’s backgrounds from Google DeepMind, OpenAI, and Boston Dynamics, along with the powerful generalization capability demonstrated by GEN-1.5—enabling robots to learn new tasks from a single demonstration without retraining. In just two months, they have become a highly anticipated new star in the embodied AI space, securing approximately $600 million (about RMB 4.308 billion) in funding.

What exactly is this company? Where does GEN-1.5’s generalization capability truly shine? What aspect of the Generalist are investors most attracted to?

You'll know after reading it:

If traditional industrial robots are like chefs who can only follow a fixed recipe, Generalist’s newly released GEN-1.5 is more like a novice apprentice who can immediately take over after watching the master chef perform once.

In the past, robots had to undergo lengthy pre-deployment development, and their problem-solving approaches generally followed a “one scenario, one strategy” pattern—writing one program to tighten screws on the production line and another entirely different program to sort parts.

Even if the task remains unchanged, any alteration in the object's position, shape, or tool may require engineers to re-collect data, adjust parameters, and retrain the model.

GEN-1.5 is trying to lower the barrier to deploying robots from “please have an engineer rebuild it” to “have an on-site operator demonstrate it once.”

In the Generalist demonstration video, staff only need to demonstrate the task once using a handheld gripper or robot; the model directly understands the task and begins operation by interpreting 3- to 12-second sensor and motion data as "physical prompts," without requiring any gradient updates or fine-tuning.

For example, it can organize actions on the fly based on demonstrations—such as unscrewing a glass jar, zipping open a pencil case, sweeping blocks into a bowl, or even using a brush and dustpan it has never seen before.

Of course, GEN-1.5 isn't yet so advanced that it can achieve perfect accuracy with just a single glance.

Official testing shows an average single-demo success rate of 59% across 10 simple, short-term tasks; with fine-tuning using approximately 5 minutes of data and 10 gradient update steps, the success rate increases to 83%.

This result isn't perfect, but it proves something more important: the task adaptation of robots has the potential to be compressed from months to seconds.

What investors may be drawn to is precisely this potential to become a universal foundation for robots.

For customers, the same model can be rapidly deployed across different factories, warehouses, and laboratories, reducing costs associated with engineering deployment, data collection, and production line calibration.

For investors, Generalists are not selling individual robots—they are selling an intelligent layer that runs across different hardware, naturally expanding the market scope.

For the Generalist themselves, these general skills can also help drive the data flywheel:

The stronger the model, the more real-world business tasks it can handle; the more deployments, the richer the physical interaction data it gathers; new data then feeds back to train the next generation of models. Ultimately, GEN-1.5 aims to enable robots to continuously learn and work.

This ability to achieve second-level adaptation instead of month-long development did not arise out of nowhere—it stems directly from the Generalist founding team’s decade-long firsthand experience with the pain points in the robotics industry.

The founding team of Generalist has brought together the two most critical capabilities in the robotics industry.

CEO Pete Florence holds a Ph.D. in Computer Science from MIT, where he studied under roboticist Russ Tedrake. He previously served as a Senior Research Scientist at Google DeepMind, where he helped pioneer vision, language, and action models and trained DeepMind’s first multimodal large model.

△

Chief Scientist Andy Zeng earned his bachelor’s degree in computer science and mathematics from the University of California, Berkeley, and later received his Ph.D. in computer science from Princeton University.

He previously served as a research scientist at Google DeepMind, where he studied robots capable of writing their own code and invented a method for大规模采集机器人数据 using handheld grippers.

△

The CTO, Andrew Barry, is responsible for turning ideas into reality within the machine.

He earned his bachelor’s degree from Olin College and his Ph.D. from MIT, with research interests in robotic control and high-speed autonomous obstacle avoidance;

Before joining Generalist, he worked at Boston Dynamics for approximately five years as a senior robotics engineer, contributing to the development of the robotic arm for the robot dog Spot.

In other words, he doesn’t just know what the model should output—he also understands how the robot should move in the real world to avoid falling, shaking, or stalling.

△

These three individuals did not form a team on a whim just because AI became popular; their combined work over the past decade+ has consistently pointed to the same bottleneck in the robotics industry:

The hardware can already run, jump, and grab—the real bottleneck to deployment is that the robot's "brain" can't keep up with its "body."

DeepMind brings the model's ability to generalize across different tasks, objects, and scenarios;

Boston Dynamics brings a sense of physicality to mechanical structures, control systems, and real-world environments.

This is also why Generalist chose to train a general-purpose brain for robots rather than betting on a specific robot body—they understand that only when both the model and hardware meet high standards can robots move from laboratory demonstrations to real-world applications in factories and warehouses.

Otherwise, no matter how elegant the model, it may only result in a few more papers; no matter how impressive the robot’s movements, it can still only perform pre-programmed routines.

However, GEN-1.5 still has significant room for improvement.

However, being able to understand it after one look doesn't mean GEN-1.5 is ready to start work in a factory.

The generalist has put the brakes on this capability. According to the official technical blog, the team is currently testing mostly simple, short-duration tasks such as twisting jar lids, zipping zippers, and picking up objects.

GEN-1.5 achieves an average single-shot success rate of 59% across 10 tasks, still showing a clear gap from the production requirement of long-term, low-fault stable operation.

Once inside the factory, robots must contend with material variations, equipment wear, human interference, and unexpected malfunctions—challenges far greater than those in laboratory demonstrations.

More importantly, "learning from one demonstration" primarily addresses task adaptation and does not automatically equate to low-cost commercialization.

Moreover, whether robots can be deployed at scale depends on hardware costs, operating speed, maintenance expenses, security, and system integration capabilities. Even if the model learns quickly, if the robotic arms are expensive, prone to frequent failures, or require extensive engineering adjustments for each deployment, the business case still won’t add up.

However, the value of GEN-1.5 should not be measured solely by its current success rate.

The emergence of GEN-1.5 demonstrates that robots have the potential to quickly understand new tasks through brief "physical prompts," just as large models understand text prompts.

Although it has not yet deployed robots on long-term production lines, it has already accelerated the speed at which robots learn new tasks—from months to seconds.

What’s more noteworthy about GEN-1.5 is the shift from developing separate programs for each task to using a single model that learns multiple tasks.

It may not yet be a qualified worker, but it at least demonstrates a potential learning approach for future general-purpose robot brains.

In this sense, GEN-1.5 is currently more like a roadmap toward a general-purpose robot than a finalized product specification.

What truly excites the market is not just what it can do today, but the potential it demonstrates for robots to continuously learn more.

If past robot companies were competing to become "specialized champions," Generalist tells investors a story about a "generalist robot."

When robotic arms being able to grasp, place, and twist is no longer remarkable, the standards for capital interest have also changed.

The market cares less about how many actions a robot has pre-learned, and more about how quickly it can learn new actions, how much data it requires, and what the deployment cost is.

Although GEN-1.5 still has a long way to go on its journey from “learn by watching once” to “consistently work eight hours a day,” Generalist at least validates a possibility: robots don’t have to be permanently confined to “one scenario, one strategy”—they can also rely on general-purpose models to adapt quickly to unfamiliar tasks, just like humans.

And as the time it takes for robots to learn new skills shrinks from months to seconds, the marginal cost approaches zero—what may truly need to be redefined is not just the robotics industry.

At that time, where will the core value of human workers lie?

This issue may be more worthy of discussion than GEN-1.5 itself.

Reference link:

[1]https://www.axios.com/2026/08/24/robotics-ai-generalist-200m

[2]https://forgeglobal.com/generalist_stock/

[3] https://generalistai.com/blog/gen-1.5

[4] https://generalistai.com/about

[5]https://www.peteflorence.com/

[6] https://andyzeng.github.io/ [7] https://abarry.org/

This article is from the WeChat public account "Quantum Bit," authored by Wen Ting.