The processing window counts the model's tokens and shows what it is writing again, instead of waiting in silence for a stage to finish.
FixedBreakdown**The processing window shows the model at work again.** For the past few days a stage showed nothing until it was done — no token count, no text, just “Waiting for the model to start streaming…” — which on a feature's first stage lasts a quarter of an hour and looks like nothing is happening. The token counter and the live text move again as the model writes, and on a long script the bar ticks forward as each scene is read. In the minute or two before its first word, while the model reads the screenplay and thinks, the window says how much it has read instead of looking stuck.