Founded three months ago, Simate's debut physical AI model takes the RoboDojo crown Simate has arrived with momentum. On September 26, Chinese tech media 量子位 (QbitAI) reported that Simate (Silicon Mate), a company founded just three months ago, has released its first general-purpose physical AI system, Simate-beta. According to the company, the model took first place on the RoboDojo leaderboard — with a base model that received no benchmark-specific tuning at all.
Simate-beta is positioned as a "general physical fast system," demonstrating memory, long-horizon task execution, fine manipulation, and task adaptation — some of the hardest problems in physical intelligence. The team has serious credentials: its core members previously pushed one-shot end-to-end driving models to Tesla FSD-level capability and shipped them into mass production. Very few teams worldwide have crossed that bar, and even fewer have taken that expertise into general physical AI. From day one, Simate set an unusual R&D course: AI for Physical AI — letting AI itself take part in researching and iterating physical intelligence.
From steering wheels to robot arms: why driving experience is coveted
Driving and physical AI may look like different trades, but underneath they share the same hard problems: making decisions in the open real world, generalizing from long-tail scenarios, and evolving through a data flywheel. The one-shot end-to-end approach to driving — perception, decision, and execution inside a single model — is nearly isomorphic to what embodied AI pursues today in end-to-end manipulation. That's why a team with FSD-grade mass-production experience draws instant attention: what they bring isn't just algorithms, but a complete engineering playbook for "making models behave in the physical world." (Related: embodied AI heads for the public exam hall.)
What does shipping in three months tell us?
The most thought-provoking part is actually the speed: three months from founding to a leaderboard-topping debut model. It suggests the physical-AI tech stack is converging fast — and validates the "AI accelerating AI research" playbook. A sober footnote is still warranted: Simate-beta remains in beta, a leaderboard score is only one lens on R&D strength, and true productization and commercialization are still ahead. Either way, competition in physical intelligence has moved from "who tells the story first" to "who ships first."
