GPT-6.2 Astra scores 41.2% on XDOF Ember-0 robotics benchmark
A doubling of benchmark score in two months points to how quickly general models are absorbing embodied capability, which may compress the differentiation window for robotics-specific AI companies. Robotics engineers and embodied AI researchers should track XDOF's benchmark releases as saturation will force new evaluation methods. Enterprise teams evaluating foundation models for robot control can use Ember-0 scores as one signal when comparing general models against specialised robotics stacks.