Google ties Antigravity rate limits to agent 'work done' and moves free users to a weekly quota

What Google actually changed

Google adjusted how it meters access to Antigravity, its agentic development platform. Google AI Pro and Ultra subscribers now get priority access with the platform's highest rate limits, and their quotas refresh every five hours.

Free-plan users are being moved onto a larger, weekly-based rate limit. Google frames this as a way to reduce how quickly free users run into limits during a single project — trading frequent refreshes for a bigger bucket that resets less often.

The company attributes the change to demand, saying the response to Antigravity has been strong and that enhanced support for subscribers is one way it is meeting that demand.

Metering by 'work done,' not by request

The most technically interesting line is how Google describes consumption. Quota is not tied to a simple count of prompts or calls.

Remember, usage is correlated with the "work done" by the agent; straightforward tasks consume less quota than complex reasoning.Montana Labs

That framing matters for anyone budgeting agentic workflows. A single instruction can trigger a small amount of work or a long chain of reasoning, and the two draw down the same quota very differently. Cost predictability now depends on the shape of the task, not the number of times you hit send.

What stays constant across tiers

Google is careful to separate access from capability. Regardless of tier, all users keep Gemini 3 Pro, unlimited tab code completions, and access to every product feature, including the Agent Manager and Browser integration.

In other words, the differentiation is throughput and refresh cadence, not the underlying model or feature set. Free users are not on a weaker model; they simply get a slower-replenishing pool of agent work.

The implication: paid tiers buy refresh speed for sustained agentic sessions

The five-hour refresh for Pro and Ultra versus a weekly cap for free users tells you what Google is charging for: the ability to run heavy, reasoning-intensive agent work repeatedly through a day without waiting.

For teams evaluating Antigravity, the practical question is not which model you get — it is the same Gemini 3 Pro everywhere — but whether your workflows lean on complex reasoning that burns quota fast. If they do, the value of a paid tier is measured in how often the meter resets, not in unlocked features.