Antigravity CLI usage limits
What the 5-hour window and the weekly limit are on Google's plans, when they reset, what Google publishes and what it does not, and how many real coding tasks fit in a week on each plan, from our W38 2026 run.
On Google AI Ultra 20x, Gemini 3.8 Flash fits about 1900 tasks a week through Antigravity CLI, estimated from the tier multiples. That is 99 passes of our fixed task set (real-world issues plus spec planning, 19 tasks) in one weekly window.Measured
Provider pages reviewed Sep 14, 2026. Figures from the W38 2026 grading run, published Sep 14, 2026.
A 5-hour refresh under a weekly limit
Google's Antigravity plans page describes Pro and Ultra quota as "refreshed every five hours until weekly limit reached", and adds that "rate limits are correlated with the amount of work done by the agent, which can differ from prompt to prompt" (antigravity.google/docs/plans, retrieved 2026-09-14). It publishes no request, token or message count for either tier.
The account's own quota summary is what our prober reads: one weekly bucket for the Gemini model group Antigravity serves, and since 2026-08-05 a 5-hour bucket beside it, both as a remaining fraction fine enough that a single short run moves it. Every Antigravity row on the leaderboard is read on the weekly bucket.
When the limits reset
Each bucket reports its own reset time and rolls over on its own clock. Google publishes no fixed weekday. On Google AI Pro the weekly limit holds exactly six fully spent 5-hour windows: every run in our sample agreed on that ratio, which is the cleanest window-to-week relation of any provider we meter.
Pro and the Ultra tiers
Google's plans page describes Pro as carrying expanded agent-model limits in Antigravity and Ultra as carrying higher ones, and sells Ultra as up to 20x the Gemini usage limits (one.google.com/about/google-ai-plans, retrieved 2026-09-14; the page served to us in a localized edition). Our plan catalog lists Google AI Ultra 5x and Ultra 20x at their sourced prices. The table scales the Google AI Pro measurement by those multiples to fill the Ultra rows, marked est. We have not run on Ultra.
Effort and model choice
Google publishes no figure for how an effort or thinking setting changes consumption. We grade Gemini 3.1 Pro at --effort high and Gemini 3.8 Flash at --effort medium, which are different models, so there is no same-model effort comparison to quote. The two columns of the table show how much the model choice alone moves the weekly meter.
Formerly Gemini CLI
The older Gemini CLI counted requests per day rather than a percent of a window: 1,500 requests a day on Google AI Pro and 2,000 on Ultra (geminicli.com/docs/resources/quota-and-pricing, retrieved 2026-09-14). Antigravity replaced it in our roster; its meter is the percent-of-window bucket described above, and the daily request counts do not apply to it.
Whether the Antigravity weekly limit is shared with the Gemini app is not stated on the pages we retrieved.
How much a week holds on Google AI Pro
Google AI Pro's weekly limit holds 6 fully spent 5h windows, out of the 33.6 a calendar week contains. Measured over 503 runs in 31 session windows, every one of which agreed.
How many tasks fit in a week
One suite run is one pass of our fixed task set: real-world issues plus spec planning, 19 tasks. Each cell is how many of those passes one weekly window of the plan fits, from the share one pass consumed of the provider's own meter. A task is a whole multi-turn agent session on a real codebase, so this reads far lower than a message count would. Rows marked est scale the Google AI Pro measurement by Google's advertised usage multiples (one.google.com, retrieved 2026-09-01), never by price; we have not run on those plans.
| Plan | Gemini 3.8 Flashoverall D+ | Gemini 3.1 Prooverall F |
|---|---|---|
| Google AI Pro$20/mo · measured | about 5 runsabout 94 tasks | about 8.2 runsabout 160 tasks |
| Google AI Ultra 5x$80/mo | about 25 runsestabout 470 tasks | about 41 runsestabout 780 tasks |
| Google AI Ultra 20x$180/mo | about 99 runsestabout 1900 tasks | about 160 runsestabout 3100 tasks |
Hover or focus a cell for the derivation. "More than" marks a run the meter could not resolve: it read under one percent for the whole pass, so the count is a floor. A share is always of the plan's own allowance and is not comparable across providers.
The plans
Google AI Pro at $20/month: the plan our subjects grade on, so its row is measured. Is Google AI Pro worth it? · scorecards: Gemini 3.8 Flash, Gemini 3.1 Pro
Google AI Ultra 5x at $80/month: not run on; its row is an estimate from the measured plan. Is Google AI Ultra 5x worth it? · scorecards: Gemini 3.8 Flash, Gemini 3.1 Pro
Google AI Ultra 20x at $180/month: not run on; its row is an estimate from the measured plan. Is Google AI Ultra 20x worth it? · scorecards: Gemini 3.8 Flash, Gemini 3.1 Pro
What Google publishes
| Figure | Status |
|---|---|
| Quota refresh interval (5 hours) | published |
| That a weekly limit exists | published |
| Tier multiple for Ultra (up to 20x) | published |
| Request, token or message count per tier | not published |
| Reset weekday or hour | not published |
| Effort or thinking multiple | not published |
| Whether the Gemini app shares the limit | not published |
| How many 5-hour windows the weekly limit holds | our meter reads it |
| Percent of each window one run consumed | our meter reads it |
Sources: Antigravity plans (antigravity.google/docs/plans), retrieved 2026-09-14; Google AI plans (one.google.com), retrieved 2026-09-14; Gemini CLI quota and pricing (geminicli.com), retrieved 2026-09-14. Our own meter readings are described on each scorecard under "How the plan cost was measured".
Frequently asked questions
How many Antigravity tasks a week does Google AI Pro give you?
In one weekly window of Google AI Pro, one pass of our fixed task set (real-world issues plus spec planning, 19 tasks) fits Gemini 3.8 Flash: about 5 passes (about 94 tasks); Gemini 3.1 Pro: about 8.2 passes (about 160 tasks). Figures are measured on Google AI Pro itself, W38 2026. A benchmark task is a whole multi-turn agent session, not a message.
How much of Google AI Pro does one Antigravity CLI task use?
About 1% of the weekly window on Gemini 3.8 Flash: one weekly window of Google AI Pro fits about 94 tasks, read off the provider's own meter before and after each graded run, W38 2026. A benchmark task is a whole multi-turn agent session, not a message, so a provider's message estimate is a different unit.
When does the Antigravity weekly limit reset?
On its own rolling clock: the account's weekly quota bucket reports a reset time, and the 5-hour bucket refreshes every five hours until the weekly limit is reached (antigravity.google/docs/plans, retrieved 2026-09-14). Google publishes no fixed weekday. On Google AI Pro the weekly limit holds exactly six fully spent 5-hour windows in our measurements.
Does a higher effort setting burn the Antigravity limit faster?
Google says its rate limits track the amount of work the agent does, which differs from prompt to prompt, and publishes no effort multiple. We grade Gemini 3.1 Pro at effort high and Gemini 3.8 Flash at effort medium, different models, so we have no same-model number to quote.
Is the Antigravity limit shared with the Gemini app?
Not stated on the pages we retrieved (antigravity.google/docs/plans and one.google.com/about/google-ai-plans, 2026-09-14). Our meter reads the weekly bucket of the Gemini model group that Antigravity serves, on a dedicated account with nothing else drawing on it.
Other harnesses: Claude Code · Codex CLI. The leaderboard shows the same usage per plan-price band; compare subscriptions puts the plans side by side.