Available variant
Prices are in USD per minute of session time, metered per second. See
Pricing & Billing for when billing starts and stops.
DreamZero predicts future video and robot actions from camera observations, measured state, and a
task instruction. Reactor runs inference and retains recent observations across predictions; your
client captures observations and controls the robot.
The DROID variant emits
action_chunk messages once a prompt and all camera streams are available.
It does not wait for a separate prediction request or a report of executed actions. Your application
checks the targets, decides which ones to execute, and replaces pending actions as new predictions
arrive. Optional predicted video can help you inspect the model’s output; it does not replace camera
observations or safety checks.
The upstream DreamZero project supplies the model family.
Matching the number of action values alone does not establish compatibility with another robot.
Start with DROID compatibility, then
get your first actions or
run in simulation.