DYNA-1 is a vision-language-action foundation model from Dyna Robotics, designed for commercial-grade autonomous dexterous manipulation at production scale. It achieves over 99 percent success rate on complex multi-step manipulation tasks in 24-hour non-stop operation. The model generalizes across environments without task-specific post-training and supports self-improvement through deployment data. It has been demonstrated on tasks including laundry folding and object manipulation in live public settings.