Startup Generalist Unveils GEN-1.5 Model: Robots Learn New Physical Tasks in 3 Seconds

US startup Generalist AI has unveiled GEN-1.5, a foundational neural network model designed to control physical robots. The system allows robotic arms to learn new actions from a single short demonstration lasting 3 to 12 seconds, without requiring additional training or reprogramming.

A human operator simply needs to perform the required movement in front of a camera. The video recording is instantly loaded into the model’s context, allowing the robot to immediately replicate the operation, much like text-based AI models process prompt instructions.

Improvising with Objects and Combining Skills

During testing, the robot learned to open jars, unzip zippers, pull bills out of wallets, and sort objects. Rather than simply copying motion trajectories, the system adapts to changing conditions and the positioning of items on a table.

According to WIRED, the model can devise alternative solutions. When researchers removed a brush during a test involving sweeping a block off a table, the robotic arm used a nearby banana instead. When a dustpan was introduced, the robot immediately figured out how to scoop up the object and deposit it into a container.

The model can also combine separate physical prompts into a single sequence: if shown how to open a wallet and then separately shown how to pull out a bill, the machine generates the missing intermediate movements itself to complete the entire task. The system can learn from human actions as well as transfer skills from virtual simulations to the real world.

Introducing GEN-1.5, a one-shot learner.

It can learn new tasks in a few seconds. Show it what to do, and it generalizes.

This capability emerged from pretraining on physical data at scale, as a step towards our mission of building general intelligence for the physical world. pic.twitter.com/ptB9ElYXMU

— Generalist (@GeneralistAI) August 19, 2026

Success Rates and Accuracy

The developers note that the technology is currently focused on basic manipulation tasks:

• After a single 3-to-12-second demonstration, the average success rate for test operations is 59%;
• After five minutes of demonstrations and ten quick fine-tuning steps, accuracy increases to 83%.

The company compares the rollout of GEN-1.5 to the emergence of one-shot learning in large language models: instead of spending hours collecting datasets, operators in warehouses and factories will only need to show the robot the required action once.

Backed by Nvidia and Jeff Bezos

According to an official company announcement, the startup raised $400 million in a funding round in June 2026, bringing its total capital raised to more than $500 million.

The list of key investors in Generalist AI includes semiconductor giant Nvidia and Amazon founder Jeff Bezos’s venture capital firm Bezos Expeditions.