The normal path sends your full prompt straight through. The TC5 path compresses the prompt first, then sends the shorter version through the same model with the same instruction stack.
Type once, then send the same request through both workers. Preview token charts refresh as you type, and live runs use the same low-variance settings on both paths.
Full prompt sent directly to grok-4-fast-non-reasoning with the shared master prompt.
The normal worker answer will appear here after the prompt is sent through the raw path.
Trimmed prompt sent to the same model with the same shared master prompt.
The TC5 worker answer will appear here after the compressed path finishes.
Preview estimate for the full prompt until you send it.
Preview estimate for the compressed prompt until you send it.
Appears after the live model call.
Appears after the live model call.
Backend preview count for the untouched user prompt before the master prompt is added.
Backend preview count for the compressed prompt that goes through the TC5 path.