Robot intelligence startup Generalist AI Inc. has unveiled GEN-1, a robot learning foundation model, SiliconANGLE reported on Sunday.

GEN-1 is a follow-up model released five months after GEN-0, which debuted last October. The company said it recorded a 99 percent success rate across many tasks and performs work at about three times the speed of current top-tier models.

The researchers focused on three areas for improvement: reliability, speed and improvisation. The company said GEN-1 goes beyond simple repetitive motions and handles complex multi-step tasks, such as folding multiple pieces of laundry and assembling items, through time and spatial reasoning. On speed, it completes box assembly in about 12.1 seconds. GEN-0 and the Physical Intelligence model pi-0 took 34 seconds for the same task.

The most notable feature is its improvisational response capability. In situations where an object slips, a lock does not work or an item becomes deformed, GEN-1 completes the task by approaching from different angles and changing its method. For example, if a shirt flips while being folded, it identifies the situation on its own and corrects it. Existing AI models often repeatedly re-evaluate in such situations or fail the task.

The researchers said not all tasks reached a 99 percent success rate. Complex tasks that make it difficult to meet practical speed and reliability at the same time in everyday environments did not meet that standard.