Simate Debuts Simate-beta and Claims Top RoboDojo Benchmark Result
The AMW Read
Simate is a new physical-AI entrant with an early benchmark claim, but the reported result and underlying system remain company-disclosed and sub-segment in scope.
Simate Debuts Simate-beta and Claims Top RoboDojo Benchmark Result
Physical-AI startup Simate said it has released Simate-beta, its first general-purpose physical “fast system” for real-time robotic perception and action. The company reported a first-place RoboDojo result as of September 23, with an average score of 33.95 and a 27.96% success rate; it said the model was not specially optimized for the benchmark. Simate described demonstrations covering task adaptation from demonstrations, memory, long-horizon execution, and fine manipulation, but has not yet disclosed the model architecture or parameter count.
The announcement is an early data point in the race to build robotics systems that can generalize beyond task-specific training. Simate is pursuing a division of labor in which a fast, deployable action model handles millisecond-level responses while larger general-reasoning models support higher-level planning. Its stated research stack combines a modular model framework, an automated experiment system, and internally built training, simulation, inference, and evaluation infrastructure. The company says this setup lets researchers run many independent experimental paths in parallel, using world models and simulation to screen candidates before hardware testing.
For builders, the important claim is not a single benchmark rank but whether automated research workflows can shorten the loop from robotic failure to new data, model changes, evaluation, and real-world validation. Investors should look for independently reproducible results, details on deployment latency and hardware performance, the gap between simulated and real-world outcomes, and evidence that Simate’s reported rapid iteration can continue as tasks become more varied. The company also said it has completed several financing rounds worth hundreds of millions of RMB each and is testing its AutoResearch platform with academic researchers.