Gmail's Smart Compose (and tab-completion generally) predicts your next words and offers them as ghost text: the system picks the moment, the position, and the content, and you accept or ignore. OpenCues inverts the relationship: you place a _ exactly where a response should land, and the system only decides what goes there.
Prediction vs request
Smart Compose answers "what will they type next?" That's useful for finishing sentences, and worthless for the things OpenCues does: computing 4 + 4 = _, fetching weather london _, rewriting a paragraph with make this formal _, or drafting a whole email. Those aren't predictions of your next words; they're requests, and a request needs an addressing scheme, which is what the _ is.
Attention economics
Ghost text charges attention every time it fires, right where you're typing, whether you wanted it or not. A blank costs nothing until you summon it, and a blank that can't resolve just stays as a visible _. The passive layer OpenCues does have (dimmed word alternatives) is deliberately non-blocking: no popup, no ghost text, fully ignorable.
One ecosystem vs anywhere
Smart Compose lives inside Google's own products (Gmail, Docs, Chat). OpenCues is one standard across every supported host, with your own model behind it.
Related: OpenCues vs autocomplete · What is a blank? · OpenCues vs Apple Writing Tools · GitHub repo