Gemini API Free Tier: Limits, Models and Gotchas (2026)

Ask five websites what the Gemini API free tier allows and you'll get five different answers. That isn't because they're careless. Google changed the limits several times over the past year, stopped printing a simple per-model table in its documentation, and moved some models in and out of the free tier. This guide tells you what's reliably true, what's disputed, and where to look for the number your own project actually gets.

Is the Gemini API free?

Yes. The Gemini API has a free tier that isn't a trial and doesn't expire. You create a key in Google AI Studio, and eligible models are free within rate limits for requests per minute, tokens per minute and requests per day. Limits apply per Google Cloud project, and your exact numbers appear in AI Studio.

What's actually free?

Here's where sources disagree, so I'll separate the solid parts from the shaky ones.

Why the numbers you read online don't match

Third-party tables circulate wildly different values: some list 15 requests per minute and 1,500 per day for Flash, others quote 5 to 15 requests per minute and 100 to 1,000 per day depending on model. Reddit users reported sharp cuts in late 2025, and some figures from that period stuck around in blog posts long after they changed.

How Gemini free-tier limits work

Three measures matter, and a fourth shows up for image models.

The privacy catch you shouldn't skip

On the free tier, Google's terms allow it to use your inputs and outputs to improve its products, and human reviewers may read them. That's the price of free. If you're sending customer data, private documents or anything you wouldn't want reviewed, don't use the free tier for it. A paid tier changes that arrangement, so read the current terms on Google's API terms page before you decide.

Free grounding allowance

Gemini 3 models include an allowance for Google Search grounding, reported at 5,000 free search-grounded prompts per month shared across those models, with a per-request charge after that. Treat this as a reported figure, and verify it on the pricing page before you rely on it.

Quick start code

# pip install google-genai
from google import genai

client = genai.Client(api_key="YOUR_AI_STUDIO_KEY")

response = client.models.generate_content(
    model="MODEL_ID_FROM_AI_STUDIO",   # copy a Flash model ID from the console
    contents="Explain rate limits in two sentences.",
)
print(response.text)

Handling 429 errors

A 429 means you've hit a limit. Wait and retry, and if the daily cap is gone, wait for the reset.

# Exponential backoff for 429 errors
import time

def ask(client, prompt, model, retries=4):
    for attempt in range(retries):
        try:
            return client.models.generate_content(model=model, contents=prompt)
        except Exception as e:
            if "429" in str(e) and attempt < retries - 1:
                time.sleep(2 ** attempt)
                continue
            raise

Backoff handles brief per-minute bursts. It can't fix a spent daily budget. If you see 429s early in the day, you're probably hitting tokens per minute, so shorten your prompts or add pauses.

Interactive guide: set up in ten minutes

Build your own: a quota tracker

Paste this into Claude or any AI chat to get a small tool that warns you before you hit the free tier.

Write a Python module called quota_guard.py for the Gemini API free tier.

Requirements:
- A class QuotaGuard(rpm, rpd, tpm) that I set from my AI Studio limits.
- A method wait_if_needed(estimated_tokens) that sleeps just long enough
  to stay under RPM and TPM, using a rolling 60-second window.
- A method record(tokens_used) that tracks requests and tokens today and
  saves the counts to a local JSON file.
- Reset the daily count at midnight Pacific time (America/Los_Angeles).
- Print a warning when I pass 80 percent of the daily limit.
- Include a short usage example and comments on every function.
- Use only the standard library.

Which free tier fits your project?

If you want a different provider to compare, the Groq API free limits guide shows Groq's per-model, organization-wide caps. For chat instead of code, the Claude free plan limits guide explains Claude's rolling window. And if you'd rather skip rate limits and keep data on your machine, the LM Studio setup guide walks through running a local model. For voice output, see the free ElevenLabs alternatives, which includes a Gemini option.

How this guide was checked

I compared several 2026 free-tier write-ups, a model lineup reference and a limits explainer, noted where they contradicted each other, and kept only the claims that most sources shared. Because Google doesn't publish a fixed table, exact numbers here are intentionally missing. Check AI Studio for yours.

Frequently asked questions

Further reading

• Gemini API rate limits documentation – Google's official explanation of usage tiers and how limits are measured.