gemini free tier limits 2026 Guide

Gemini Free Tier Limits in 2026: What You Actually Get (and Where It Runs Out)#

Google's free Gemini access changed more in 2026 than in any year since launch. If you last checked the limits in 2025, the rules you remember no longer apply. Free users lost model choice in the Gemini app, developers lost published rate-limit tables, and the line between "free" and "paid" moved in ways that catch people mid-project.

This guide breaks down the Gemini free tier limits as they stand in October 2026, separates the consumer app from the developer API, and shows you how to plan real work around the caps instead of discovering them at 2 a.m. with a failed batch job.

First, Know Which Free Tier You're On#

"Gemini free tier" describes two different products with two different rulebooks:

• Gemini Apps — the chat interface at gemini.google.com, plus Gemini in Search, Gmail, and mobile. Limits are compute-based and tied to your Google account.
• Gemini API — the developer platform accessed through Google AI Studio and API keys. Limits are per project, measured in requests and tokens.

Mixing these up is the most common source of bad advice online. A generous API quota tells you nothing about how many prompts you get in the app, and vice versa.

Gemini App Free Limits: The October 2026 Reset#

One model, no picker#

Since October 9, 2026, free accounts no longer choose a model. Free users get Gemini 3.5 Flash-Lite. Gemini 3.6 Flash and Gemini 3.1 Pro moved behind Google AI Plus and Pro subscriptions. The model picker that used to let you trade speed for reasoning is gone unless you pay.

That matters more than it sounds. Flash-Lite is genuinely fast and handles translation, rewriting, summarization, and routine coding help well. It is also the first model to struggle with multi-step math, long logical chains, and dense document analysis. If your task needs heavy reasoning, the free app is now the wrong tool — not because of quotas, but because of capability.

Compute-based caps, not prompt counts#

Google retired the old "X prompts per day" model in May 2026. Free usage is now metered by compute: prompt complexity, the features you invoke, and conversation length all draw from the same pool. Your allowance refreshes every five hours until you hit a weekly ceiling.

In practice, this means ten short questions may cost less than one Deep Research run or one long image-generation session. Google surfaces your remaining allowance at gemini.google.com/usage, and the app warns you before you hit the wall. Check that page before starting anything ambitious.

The 32,000-token context window#

Free accounts cap context at 32,000 tokens — roughly 50 pages of text, or about 24,000 words. Paid tiers jump to 128,000 tokens on AI Plus and 1 million on AI Pro.

Uploads count against that window. Free users can upload video up to 5 minutes and audio up to 10 minutes, versus 1 hour and 3 hours on AI Pro. Notebooks are capped at 100 with 50 sources each. Try to analyze a 200-page PDF on a free account and Gemini will quietly lose the middle of the document, then answer confidently from what it retained.

Actionable tip: Split long documents into labeled sections and ask one question per section. You get better answers than one giant upload, and you stay inside the context budget.

Gemini API Free Tier Limits: What Developers Get#

Which models are free#

The API free tier still exists and requires no credit card. As of October 2026, free input and output are available on Gemini 3.8 Flash, 3.6 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite, Gemma 4, and several audio and embedding models.

Not free: Gemini 4 Argon, Gemini 3.1 Pro Preview, Nano Banana image models, and video models. Google explicitly notes that Argon "is not available on the Free tier."

Watch out for the 2.5 family. Those models still show "free of charge," but Google is limiting access to users who actively used them in the past. New projects should start on 3.8 Flash or 3.5 Flash-Lite.

The numbers moved into AI Studio#

Google stopped publishing per-model RPM, TPM, and RPD tables in 2026. The rate-limits page now sends you to the rate-limit view in Google AI Studio, which shows the live limits for your specific project. Any blog post quoting exact free-tier numbers is either outdated or guessing.

What the documentation still guarantees:

• Three dimensions apply. Requests per minute (RPM), input tokens per minute (TPM), and requests per day (RPD). Exceeding any single one fails the request — 21 calls in a minute breaks a 20 RPM limit even if your token usage is trivial.
• Limits are per project, not per key. Five API keys in one project share one allowance.
• RPD resets at midnight Pacific time, not 24 hours after your first call.
• Preview models are tighter, and Google states plainly that specified limits "are not guaranteed and actual capacity may vary."

What the free API actually costs you#

Free API quota is paid for with data. Google's terms treat unpaid quota as an Unpaid Service and state that submitted content and generated responses may be used to improve Google products, with human reviewers able to read and annotate inputs and outputs.

Keep customer records, credentials, and anything under NDA off a free-tier key. In the EEA, Switzerland, and the UK, you may only use paid services when shipping API clients to end users — free keys are for your own testing there.

Also note: since March 2026, Gemini API usage is excluded from the $300 Google Cloud Free Trial credit. Prepay credits expire after 12 months, and if your balance hits zero, every key on that billing account stops at once with HTTP 402.

Free vs Paid: A Quick Comparison#

	Free app	Free API	AI Plus ($5)	AI Pro ($20)
Models	Flash-Lite only	Selected Flash, Flash-Lite, Gemma	Adds Flash	Adds Pro, Argon access
Context	32K tokens	Model-dependent	128K tokens	1M tokens
Usage	Standard compute cap	Per-project RPM/TPM/RPD	2x standard	4x standard
Data used for training	Yes	Yes	No	No

Practical Examples: Working Inside the Limits#

Example 1: A 40-post rewrite batch#

Suppose you want to rewrite 40 blog posts through the free API. Each post averages 1,200 words, or roughly 1,600 tokens input and 1,800 output.

At 40 requests total, RPM is not your constraint — a project limit of even 10 RPM handles it in four minutes. TPM is the real risk if you send long system prompts alongside each post. Trim instructions to under 300 tokens, and you cut input load by nearly 20%.

Before committing, open AI Studio, note your RPD figure, and divide it by your daily job size. If your project shows 250 RPD, a 40-post batch leaves room for about six reruns. If it shows 50, split the batch across days.

Example 2: Handling 429 errors correctly#

A free-tier 429 comes with a code that tells you what to do. Per-minute codes (rate_limit_exceeded, too_many_requests) mean back off and retry — wait 1, 2, 4, then 8 seconds with random jitter, and cap retries at four. Daily codes (quota_exceeded) mean stop; retrying will not help until midnight Pacific.

The official SDKs retry transient errors automatically, but they will not save you from a daily cap. Log the error code, not just the status, so you can tell the two apart.

Example 3: Building a fallback chain#

Route each task by difficulty rather than sending everything to the strongest model. Use Flash-Lite for classification, extraction, and formatting. Escalate to Flash only when output fails a validation check. On the API, this keeps you inside free quota far longer than a flat model choice, and it usually cuts latency too.

Common Mistakes That Waste Your Quota#

• Creating new API keys to "reset" limits. Limits are per project. New keys share the same pool.
• Sending failed requests casually. A 400 or 500 isn't billed, but it still counts against quota.
• Designing retries for daily limits. Exponential backoff fixes per-minute throttling, not a daily ceiling.
• Assuming tomorrow starts at your local midnight. It starts at midnight Pacific.
• Treating the free app and free API as one budget. They are separate systems with separate rules.

The Bottom Line#

The Gemini free tier limits in 2026 are tighter and less predictable than they were a year ago, but they are still usable — for the right jobs. The free app is a fast Flash-Lite assistant with a 32,000-token window and a compute cap that rewards short, focused sessions. The free API is a genuine development sandbox for Flash and Flash-Lite models, provided you keep sensitive data out and check your project's live limits in AI Studio rather than trusting a table from 2025.

Start by measuring your actual per-request token cost, confirm your RPD in AI Studio, and route heavy reasoning tasks to a paid tier only when you truly need them. Do that, and the free tier will carry more of your workload than most people assume.

---

About the author: AI Pulse Daily is written by practitioners who use these tools daily. We never recommend anything we have not personally tested. Affiliate disclosure.

FAQ#

What are the Gemini free tier limits for 2026?#

In 2026, the Gemini free tier offers up to 15,000 API calls per month, a maximum of 2 GB of data processed per request, and 100 GB of total data transfer. The tier also includes access to the base language model with a rate limit of 5 requests per second.

Can I use Gemini's free tier for commercial projects in 2026?#

Yes, the Gemini free tier allows commercial use as long as you stay within the monthly limits (15,000 calls, 2 GB per request, 100 GB total transfer). If you exceed these limits, you’ll need to upgrade to a paid plan to avoid service interruptions.

How does Gemini's free tier compare to other AI tools in 2026?#

Compared to competitors, Gemini’s free tier provides a higher request quota (15,000 calls) than OpenAI’s free tier (5,000 calls) but a lower per‑request token limit (2 GB vs. 4 GB). It also includes built‑in content moderation at no extra cost, which many other platforms charge for.

What happens if I exceed Gemini's free tier limits in 2026?#

When you exceed any of the free tier limits, Gemini will temporarily throttle your requests and return a 429 error. You can continue using the service by upgrading to a paid plan or waiting until the next billing cycle when the limits reset.