Gemini 4 Argon usage limits: how long AI Pro and Ultra last in the Gemini app, Antigravity and AI Studio

Each surface has its own meter and none publishes a count. Google One help says "Each product has its own AI usage limits", and BreezeOg (@BreezeOg1) reports that Gemini app use does not draw down Antigravity quota.

Table: What Google officially says, by surface (as of 18 October 2026)

Sample/method: Sources: Google's Gemini help center, Antigravity docs (Plans, Models), the AI Studio Google AI plans page and Google's launch blog, all opened on 18 October 2026.

Gemini app. Limits are compute-based: prompt complexity, the models and features used, and chat length all count. "Your limit refreshes every 5 hours until you reach your weekly limit." Pro gets 4x the standard limits and Ultra 5x or 20x Pro. The help page does not name Argon.

Antigravity. Argon on Pro is for paid, non-trial subscriptions only, and only in the Antigravity app, CLI and IDE extensions, not the standalone Antigravity IDE (models page). Since 16 October 06:52 UTC+8, Pro has a separate weekly-only Argon quota on top of the Gemini bucket for 3.8 Flash (official @antigravity post). Ultra's quota refreshes every five hours under the highest weekly cap. Google says usage "correlates with the amount of work done by the agent".

AI Studio. A daily quota that resets at fixed times, not a rolling window. The plans page (updated 16 October UTC) lists Argon for Pro and Ultra; the Google AI Studio account announced Pro access to Build and Playground on 17 October 07:39 UTC+8. After the daily quota you need a Gemini API key with billing.

Prices. AI Pro is $19.99 a month, Ultra 5x $99.99 and Ultra 20x $199.99 (US). The API costs $2 per million input tokens and $10 per million output until 30 November 2026, then $4 and $20. Pro includes $10 and Ultra $40 or $100 of monthly Google Cloud credits.

A number posted on 14 October and one posted on 17 October may describe different rules.

Table: How the rules changed in one week (UTC+8)

Sample/method: Times converted from post timestamps to UTC+8; official pages checked on 18 October 2026.

At launch, Antigravity and AI Studio plan access were Ultra-only, so most Pro posts from 14–15 October complain about being locked out, not about quota.

The 17 October event was a one-off refill: Varun Mohan (@_mohansolo) posted "Resetting Argon and Gemini model quotas for everyone". Google did not say the limits changed. Afterwards Leo (@leo114119) passed on others' reports that Ultra 20x Argon quota in Antigravity was 75–90% higher, while Hakm (@hakmgpt) said the Pro app allowance felt at least 75% smaller than at launch. Neither is verified, and they point in opposite directions.

In these reports, a Pro user's 5-hour allowance in the Gemini app covers roughly 4 to 10 Argon prompts depending on the task, and the 5-hour meter runs out long before the weekly one.

Table: Pro users in the Gemini app (self-reports, not verified)

Sample/method: Sample: posts among the 50 first-hand reports that state the Pro plan, the Gemini app and a concrete number; figures are the authors' own, some with screenshots, not re-tested by us.

Hakm's two screenshots: on Low, "pick a number from 1 to 10" moved 5-hour usage from 0% to 2%, with the weekly meter still at 0%. (Item: Argon Low eats 2% of 5-hour quota on a single number pick)

兴雨's screenshot: the 5-hour meter at 100% while only 16% of the weekly limit was used. (Item: Argon's five-hour quota drains too fast, user complains)

兴雨 (@YxzRainy) said each plain-text message took about 15% and the meter locked after six messages. MJJ百科 (@mjjwiki) found one full 5-hour window cost 5% of the week, which by that author's arithmetic means about 20 windows a week. When the 5-hour meter is full, Pro users can carry on with Flash-Lite (Google help page), which is why 兴雨 called Argon on Pro "a one-hour trial card".

Heavy inputs cost more, as Google's help page says: Vipin (@vipinxtwt) reached 91% with a roughly 20-page PDF and four prompts.

Vipin's screenshot shows current usage at 91%. (Item: Gemini 4 Argon burns through usage limits, users complain)

James (@james_ios) hit the 5-hour limit after four prompts; Anton P. (@antonpme) was rate-limited after five or six short writing prompts and said earlier Gemini Pro models never did that.

On paid Pro, Argon in Antigravity is a single weekly bucket with no 5-hour refill, and users report draining it in roughly one to three hours of agent work, depending on thinking level.

Jeudry Peña's screenshot shows the panel layout: Gemini models and Claude/GPT models each have weekly and 5-hour meters, while Gemini 4 Argon has a single weekly meter (here showing 46%, resetting in 6 days 19 hours). (Item: Gemini 4 Argon High uses about 1% of the $20 Pro weekly quota per minute, test finds)

Table: Reports from Antigravity (self-reports, not verified)

Sample/method: Sample: posts among the 50 first-hand reports that name Antigravity and give a number, around the 17 October reset; figures are the authors' own, not re-tested by us.

Thinking level made the biggest difference. Jeudry Peña (@JeudryPena28376) burned about 1% per minute on High and said a 50-minute game build used 56%. Jerry Xiong (@Jerry_Galaxy) stayed on Low and still had 65% after about 23 model calls, including reading two repos. Iam_ (@SPAC89) was positive: 15 minutes without sub-agents cost about 10%, or 2–3 hours a week by that author's estimate.

Jerry Xiong's screenshot: Argon weekly quota at 65% remaining, Gemini models at 99%. (Item: Gemini Argon quota test in Antigravity: 65% left after ~23 requests)

Estimates in tokens and dollars: 麦麦提 (@shengtang135754) measured 29.64 million tokens for a full Pro week at Medium; Leo put the Pro week at about $10 of API-equivalent usage. Dev Arora (@Dev_Arora_) estimated, in preliminary coding tests, that Opus 5.5 on Claude Pro gives about 17x and GPT-6 Astra on ChatGPT Plus about 7x Argon's weekly usage on AI Pro. Each is one user's estimate.

Ultra: Derya Unutmaz (@DeryaTR_) ran out of the weekly quota in under two days at extra high to max, while erukiti (@erukiti, Ultra 5x, High) said the weekly limit barely moved. Leo's launch figures for Ultra 20x ($70 per 5 hours, $220 per week in API terms) are secondhand reports, not measurements.

Google meters by compute in the app and by agent work in Antigravity, not by message count, so a model that thinks and writes at length uses more allowance per prompt.

Google's help page lists long conversations (the history is re-read each turn), large files and Deep Research as costlier than a simple chat. Antigravity's docs say complex tasks consume more quota.

Output length is one likely reason. aGim (@aGim_asf), citing Artificial Analysis, says Argon used 62k output tokens per task against 27k for GPT-6 Astra; we did not verify this figure.

StatsWire (@StatsWire) relayed unnamed "early developer telemetry" for Pro in Antigravity: about 1% of the week per prompt on Low, 6–21% on Medium and about 33% on High (secondhand).

Sample/method: Sample: first-hand, relay, compilation and complaint posts from 14 October to 18 October 01:10 2026 (UTC+8) among 239 keyword hits; each read and classified by the editor. Counts are discussion volume in the collection.

Editor's reading: the separate Pro bucket exists, in Varun Mohan's words, so Argon "doesn't eat up your Flash quota". The trade-off is that once it is empty, Pro users may wait up to a week.

Watch the right meter, keep chats short, use Low thinking for routine steps and hand execution to Flash.

• Check the meter. Gemini app: Settings > Usage Limits. Antigravity: View Usage in the model selector, or /usage (alias /quota) in the CLI.

• Start a new chat for a new topic (Google's own tip) and attach whole PDFs only when needed.

• Use Low or Medium thinking in Antigravity for routine work; in the reports above, High drained the quota much faster.

• Plan with Argon, execute with Flash (user tips, not tested by us). 麦麦提 said making 3.8 Flash the sub-agent cut a portfolio site to 7% of the quota, against at least 20% for similar work the day before. Jerry Xiong advises the same handover.

• When it runs out: the app offers Flash-Lite; Antigravity can draw on purchased AI credits if the overage setting is Always (Never waits for the refresh); AI Studio continues with a paid API key, and Pro's $10 monthly Cloud credit can cover API calls.

• Ultra? Officially 5x or 20x Pro in Gemini and Antigravity. In reports, Ultra users still split between "barely moves" and "gone in two days", so decide by how much agent work you run.

Sample/method: Sample: 239 keyword hits among 5,027 posts in the gemini-4-pro collection (17 Sep to 18 Oct 2026, UTC+8); each read and classified by the editor, AI labels used only as a cross-check.

Of 239 posts matching limit keywords, only 50 were first-hand. We excluded 17 promotions, including nine near-identical posts offering "free unlimited" Argon through one aggregator.