GEN-1
A large multimodal embodied foundation model and system that emits actions in real time. GEN-1 is trained from scratch on more than 500,000 hours of real-world physical interaction data and is designed to master physical tasks through reliability, speed, and improvisational intelligence.
Primary use case: General-purpose physical task automation, including automotive parts kitting, T-shirt and box folding, block and phone packing, and robot vacuum servicing.
Specifications
Physical
Locomotion
Wheeled
Manipulation
Hand Type
Dexterous robot hands with bimanual in-hand regrasping capability
AI & Compute
AI Capabilities
Real-time multimodal action generation, Reliable task execution with up to 99% success rates, Fast task completion at up to approximately 3x prior state of the art, Improvisational recovery from unexpected scenarios, Physical commonsense, Data-efficient adaptation to new tasks and robot embodiments using approximately one hour of robot data, Learning from experience through reinforcement learning, Harmonic Reasoning for accelerated inference
Commercial
Availability
Available to early access partners starting April 2, 2026.
Primary Use Case
General-purpose physical task automation, including automotive parts kitting, T-shirt and box folding, block and phone packing, and robot vacuum servicing.
Related news
Humanoid Robots Can Outrun Usain Bolt, But Complex Tasks Are Taking Longer
A humanoid robot beat Usain Bolt's 100-meter sprint record at the World Humanoid Games in Beijing, but while robots excel at athletic feats, they still struggle with complex tasks like folding laundry or working factory lines. JPMorgan estimates supply chain-ready humanoid robots will cost about $120,000 each, though Elon Musk targets $20,000-$30,000 for Tesla's Optimus bots, with Chinese startups like XPeng potentially beating him to market.
Humanoid Robots Can Outrun Usain Bolt, But Complex Tasks Are Taking Longer
A humanoid robot beat Usain Bolt's 100-meter sprint record at the World Humanoid Games in Beijing, but while robots excel at athletic feats, complex tasks like laundry folding and factory work remain challenging. JPMorgan analysts estimate supply chain-ready humanoid robots will cost about $120,000 each, though Elon Musk targets $20,000-$30,000 for Tesla's Optimus bots, with Chinese startups like XPeng potentially beating him to that price point.
Rhoda Tests Whether Scaling Web-Video Pretraining Makes Robots Better at Real Work
Rhoda AI released a study on September 10, 2026 testing whether scaling web-video pretraining improves robot performance, finding that larger models achieved up to 84.7% completion rates on bearing unpacking and waste sorting tasks. The company's Direct Video-Action system predicts future video frames and translates them into robot movements, with experiments showing improved performance correlating with better video prediction and increased pretraining compute.