Google AI Studio rate limits 2026: Gemini API tiers, spend caps and 429 fixes

Updated: November 4, 2026

Google AI Studio has no single daily limit. Every Gemini API project is limited on several axes at once, and a 429 error appears as soon as any one of them is exceeded: requests per minute (RPM), input tokens per minute (TPM), requests per day (RPD), plus a rolling 10-minute spend-rate limit of $10 on Tier 1, $50 on Tier 2 and $200 on Tier 3 for paid projects. Limits are applied per project, never per API key, and daily request quotas reset at midnight Pacific time.

The tier a project sits in decides most of these numbers. Free projects get model-specific free-tier quotas, Tier 1 starts as soon as an active billing account is linked, Tier 2 needs $100 of cumulative paid spend and 3 days since the first successful payment, and Tier 3 needs $1,000 and 30 days. Each billing tier also carries a monthly spend cap ($250, $2,000, and $20,000 to $100,000+), and reaching it pauses every linked project until the first day of the next month. The tables below summarize the tiers, the error codes and the most common causes of a 429 on a paid project, followed by the steps to check and fix each one.

........

Usage tier	How a project qualifies	Spend-rate limit (rolling 10 min)	Monthly billing-tier cap
Free	Active project or free trial	Not applicable	Not applicable
Tier 1	Active billing account linked	$10	$250
Tier 2	$100 paid + 3 days since first payment	$50	$2,000
Tier 3	$1,000 paid + 30 days since first payment	$200	$20,000 to $100,000+

........

··········

HOW GEMINI API RATE LIMITS ARE MEASURED IN GOOGLE AI STUDIO.

Google documents three standard dimensions for the Gemini API: requests per minute, input tokens per minute and requests per day. Each one is checked independently, so a project that sends few requests can still hit the TPM ceiling with very long prompts, and a project with small prompts can hit RPM during a burst. Some models add their own measures: image-generating models are counted in images per minute (IPM), and some models carry a tokens-per-day (TPD) limit.

Built-in tools are metered separately. On the Gemini Developer API, Grounding with Google Search on Gemini 3 and newer models defaults to 1,500 requests per day per project. Experimental and preview models have stricter limits than stable ones, which explains why the same code can run smoothly on a stable Flash model and fail quickly on a preview model ID.

Google states that the published limits are not guaranteed and that actual capacity can vary. The exact per-model values for a given project are shown on the rate limit page inside Google AI Studio, which updates automatically when the project changes tier.

........

Limit	What it measures	Typical cause of the error
RPM	Requests per minute	Too many calls in a short burst
TPM	Input tokens per minute	Long prompts or large files sent too often
RPD	Requests per day (resets midnight Pacific)	Daily allocation used up
IPM / TPD	Images per minute or tokens per day on some models	Image generation or high daily volume
Spend rate	Dollar cost over a rolling 10 minutes	Expensive requests on a low paid tier
Monthly tier cap	Total spend on the billing account per month	Account reached $250, $2,000 or its Tier 3 cap

........

··········

GEMINI API FREE TIER LIMITS AND WHICH MODELS INCLUDE THEM.

The free tier is model-specific. On the official Gemini Developer API pricing page checked today, Gemini 3.8 Flash, Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, Gemini 3.1 Flash-Lite and the Live and transcription models are listed as free of charge on the free tier, each with its own quota. Gemini 4 Argon, Gemini 3.1 Pro Preview, Gemini Omni Flash and the Nano Banana image models are not available on the free tier, so calls to them from a project without billing fail regardless of volume.

A frequent source of confusion is the 403 PERMISSION_DENIED error with a "free tier restricted" message. Google's troubleshooting guide says that creating a new API key in the same project does not fix it, because limits belong to the project, and that the remedy is to link an active Cloud Billing account. Upgrading to a paid tier does not give the project back any free-tier quota. For a broader view of what stays free, see our guide to Google AI Studio pricing and free access.

··········

⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣

GEMINI API PAID TIERS, PREPAY CREDITS AND SPEND CAPS.

Tiers, rate limits and caps are set at the billing account level, and every project linked to that account inherits them. Moving a project to a different billing account therefore changes its tier, and all API keys in a project share the same project cap and the same account total. Qualification for Tier 2 and Tier 3 counts cumulative Google Cloud spend on the linked billing account. Upgrades are automatic once the criteria are met: Free to Tier 1 is typically immediate, and later upgrades usually take effect within about 10 minutes, although Google says an upgrade can occasionally be denied.

New billing accounts default to Prepay. The minimum purchase is $5 and the maximum prepayment is $5,000; unused credits expire 12 months after purchase. When the Prepay balance reaches zero, every API key in every project linked to that billing account stops working at once and requests fail with HTTP 402 Payment Required until credits are added. Projects are not moved back to the free tier automatically. Auto-reload can top up the balance, within a monthly auto-charge limit. Switching from Prepay to Postpay is not supported, while Postpay accounts can move to Prepay on the AI Studio Billing page.

Two kinds of cap sit on top of the rate limits. The billing-tier cap is monthly and account-wide: when the account total reaches $250 on Tier 1, $2,000 on Tier 2 or its Tier 3 cap, service pauses for all linked projects until the 1st of the next month, and an increase can be requested through Google's form. The project spend cap is an experimental, optional monthly limit that owners, editors or admins can set per project on the AI Studio Spend page; Google warns that usage can overrun it by up to about 10 minutes. Spend caps are not available for invoiced (offline) billing accounts. Our guide to Google AI Studio billing, free tier and API credits covers the billing setup in more detail.

··········

GEMINI API ERROR CODES: 429, 402, 403 AND 400 COMPARED.

Google's API error reference splits rate errors into three 429 variants and lists separate codes for billing problems. Reading the exact status in the response body is the fastest way to tell which limit was reached.

........

HTTP code	Status in the response	Meaning	What to do
429	rate_limit_exceeded	Per-minute or per-second request or token limit exceeded	Wait and retry with exponential backoff
429	quota_exceeded	Daily quota exceeded	Wait for the reset or request an increase
429	too_many_requests	Too many requests in a short period	Retry with exponential backoff
429	RESOURCE_EXHAUSTED (spend rate)	Rolling 10-minute spend limit reached	Wait, send cheaper requests, or request an increase
402	payment_required	Prepay credit balance depleted	Add credits or enable auto-reload; retrying will not help
403	permission_denied	Key lacks permission, project restricted, or billing or verification issue	Check key, verification and billing status
400	failed_precondition	A prerequisite such as billing is not met	Check the project's billing status

........

Google's troubleshooting guide classifies 429 and 503 as retryable with exponential backoff: start with a wait of about one second, double it on each attempt (2, 4, 8 seconds), add random jitter so many clients do not retry together, and set a maximum number of retries. The official Python SDK already retries up to four times, with an initial delay of about one second and a maximum of 60 seconds. Errors 400, 402 and 403 should not be retried, because the request will keep failing until the underlying setting is fixed.

··········

[[ADPLUS_300x250_1]]

WHY A PAID GEMINI API PROJECT STILL GETS 429 RESOURCE_EXHAUSTED.

A 429 on a project with billing enabled usually comes from one of a few causes, each documented on Google's rate limit and billing pages. The first is the spend-rate limit: on Tier 1, only $10 of usage is allowed in any 10-minute window, so a burst of expensive requests can be refused while RPM and TPM still show spare capacity. The second is a per-model quota: limits differ by model, and preview models such as Gemini 3.1 Pro Preview have tighter ceilings than stable models.

The third cause is billing that is not attached where the traffic runs. Because limits belong to the project, an API key created in a different project from the one linked to billing keeps that other project's tier, which may still be Free. The fourth is a tier that has not yet been upgraded: Tier 2 needs both $100 of paid spend and 3 days since the first successful payment, so heavy traffic in the first days after setup stays on Tier 1 limits. The fifth is the monthly billing-tier cap, which pauses all linked projects until the next month once it is reached. Priority inference requests are a smaller sixth case: they have their own limits, by default 0.3x the standard rate limit for each model and tier, and count toward the overall interactive limits.

........

Cause of the 429	How to check it	Fix
Spend-rate limit ($10 / $50 / $200 per 10 min)	Cost of recent requests in a 10-minute window	Spread traffic, shorten context and outputs, or request an increase
Per-model or preview-model quota	Model row on the AI Studio rate limit page	Switch to a stable model or spread calls over time
API key in a project without billing	Project shown next to the key on the AI Studio API keys page	Use a key from the billed project or link billing to that project
Tier not yet upgraded	Tier shown on the AI Studio Projects page	Wait for $100 spend + 3 days (Tier 2) or $1,000 + 30 days (Tier 3)
Monthly billing-tier cap reached	Spend page and account total for the month	Wait for the 1st of the month or request a higher cap
Daily quota (RPD) used up	Daily usage on the rate limit page	Wait for the midnight Pacific reset

........

··········

HOW TO CHECK AND RAISE YOUR GOOGLE AI STUDIO RATE LIMITS.

The rate limit page in Google AI Studio shows the active limits for each model in the project, and the Projects page shows the project's current tier. Checking both before debugging code avoids the most common mistake, which is comparing a project's behavior with a quota table published for a different tier. A successful call to the models.list endpoint only confirms that the API key exists; Google notes it does not prove that inference calls are enabled for the project.

To move up a tier, billing must first be set up in AI Studio, after which upgrades happen automatically when the spend and time conditions are met. For a paid-tier rate limit increase beyond the default values, Google provides a request form linked from the rate limit page, without a guaranteed approval. Cost data can lag behind usage: AI Studio cost graphs can take up to 24 hours to update, credit drawdown usually takes minutes, and bank-transfer payments can take several days to clear before service resumes.

Batch jobs follow separate limits. The Batch API allows 100 concurrent batch requests, a 2GB input file size and 20GB of file storage, and caps the tokens queued per model across all active batch jobs; for most text models that cap is in the range of 3 million to 10 million tokens on Tier 1 and 400 million to 1 billion on Tier 2. Moving non-urgent work to batch keeps it out of the interactive RPM and TPM budget. Current per-token prices for each model are listed in our Gemini API pricing guide.

··········

⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁤

HOW FAST THE TIER 1 SPEND-RATE LIMIT IS REACHED: A COST EXAMPLE.

The spend-rate limit is easy to underestimate because it depends on price per request, not on request count. The example below assumes a request with 20,000 input tokens and 2,000 output tokens, priced at the standard text rates for prompts up to 200,000 tokens shown on Google's pricing page today, and computes how many such requests fit in Tier 1's $10 per 10 minutes and in its $250 monthly cap.

........

Model and price period	Cost per request	Requests per 10 min on Tier 1 ($10)	Average RPM sustained	Requests per month at the $250 cap
Gemini 3.5 Flash-Lite ($0.30 / $2.50)	$0.011	909	91	22,727
Gemini 3.8 Flash until Dec 31 ($0.75 / $3.75)	$0.0225	444	44	11,111
Gemini 3.8 Flash from Jan 1, 2027 ($1.50 / $7.50)	$0.045	222	22	5,555
Gemini 4 Argon until Nov 30 ($2 / $10)	$0.06	166	17	4,166
Gemini 4 Argon from Dec 1 ($4 / $20)	$0.12	83	8	2,083

........

The figures show that a Tier 1 project running Gemini 4 Argon after the December 1 price change can sustain only about eight such requests per minute before the spend-rate limit returns 429, far below most RPM ceilings. On Tier 2 the same workload gets five times the room ($50 per 10 minutes, about 416 Argon requests), which is why reaching $100 of paid spend and waiting the 3 days often resolves recurring 429 errors faster than any code change. Trimming context has a direct effect too: halving the input tokens on Gemini 3.8 Flash cuts the cost per request from $0.0225 to $0.015 and raises the 10-minute allowance from 444 to 666 requests.

··········

GOOGLE AI STUDIO RATE LIMITS: WHAT IS DOCUMENTED AND WHAT IS NOT.

Google documents the structure precisely: RPM, TPM and RPD per model and per project, a midnight Pacific reset for daily quotas, three paid tiers with fixed qualification rules, a rolling 10-minute spend-rate limit of $10, $50 and $200, monthly billing-tier caps of $250, $2,000 and $20,000 to $100,000+, and HTTP 402 when Prepay credits run out. The exact per-model RPM and TPM values are shown only inside each project's AI Studio rate limit page and can change, so any quota table copied from another account or an older article is not proof of what applies to a given project.

Two price changes on the official pricing page will tighten the spend-rate arithmetic for many projects: Gemini 4 Argon doubles to $4 input and $20 output per million tokens on December 1, 2026, and Gemini 3.8 Flash and 3.6 Flash double to $1.50 and $7.50 on January 1, 2027. Projects that already run close to their Tier 1 spend-rate limit will see 429 errors earlier after those dates unless they move to a higher tier, shorten their requests or shift traffic to cheaper models.

··········

FOLLOW US FOR MORE.

·····

DATA STUDIOS

·····

datastudios.org