Alongside the AI Pro and Ultra repricing, at I/O 2026 Google changed the metric it uses to measure Gemini app usage. It looks like a technical detail but it rewrites the daily experience of millions of users.
The end of the prompt as a unit
Until now the limit was a number of prompts per day. It looked simple but was unfair: a twenty-word prompt counted the same as a request to generate a one-minute video.
The new model is compute-used: every interaction is weighted by the complexity of the prompt, the features it activates (deep research, code execution, video generation) and the length of the conversation. The quota is no longer daily but refreshes every five hours, up to a weekly ceiling.
The smart fallback
The most interesting novelty is what happens when you blow through the cap on the more powerful models. Google doesn't block you: it automatically reroutes you to the smaller, faster models so the conversation can continue. It's an implicit admission that blocking a user mid-task is a terrible experience, and that today the gap between Gemini 3.5 Flash and the smaller models is small enough that the fallback isn't humiliating.
For AI Pro and Ultra plans, Google also introduces pay-as-you-go credits to top up if you exceed the cap: today on Antigravity and Flow, soon on the Gemini app.
Why it matters
For power users — developers, knowledge workers, researchers — it means usage planning shifts: no longer "how many prompts do I have left today" but "how much compute do I have left this week". It's a more honest metric versus real resource consumption, but it also makes prediction more opaque: a prompt that cost little yesterday may cost more today if you flip on one extra feature.
It will be interesting to see how Google communicates the metric to the average user. For now, the signal is clear: the era of the prompt as a unit of measure is over.