Gemini API Pricing Free Tier 2026: Limits and Cost Math

By Per Million · Published October 22, 2026 · 18 min read

• Free Tier
• LLM Pricing

The Gemini API free tier costs nothing. It works through Google AI Studio and an API key, needs no billing account, and gives free input and output tokens on a set list of models, including Gemini 3.8 Flash, Gemini 3.5 Flash-Lite and Gemini 2.5 Pro, up to per-project limits on requests per minute (RPM), tokens per minute (TPM) and requests per day (RPD) (pricing, billing, as of October 2026). You only start paying after you link a billing account to a project. Without one, the free tier never sends a bill. Requests over a cap get a 429 error. The exact caps for your project are shown in AI Studio, not in the public docs (rate limits).

Which option is right for you?

If you're prototyping, the free tier is enough. If you're shipping to real users, especially in Europe or the UK, set up a paid Tier 1 project from the start.

Why each pick: Flash-Lite is the cheapest paid text model on the list, so a free-tier prototype carries over to paid billing with the fewest surprises. Flash is the stronger default if quality matters more than unit cost. EEA/UK apps need paid service because of the terms, not because of cost. Batch is the easiest 50% cut for work that isn't time-sensitive.

What Is the Gemini API Free Tier?

The Gemini API free tier is a no-cost usage level, reached through Google AI Studio and an API key, that lets you call specific Gemini models up to fixed per-minute, per-token and per-day limits. No billing applies. Google's own wording on the pricing page is "Limited access to certain models" and "Free input & output tokens" (pricing, as of October 2026).

The free tier includes:

• Free text and multimodal models. The pricing page (last updated October 19, 2026) lists "Free of charge" under the free-tier column for Gemini 3.8 Flash, Gemini 3.6 Flash, Gemini 3 Flash Preview, Gemini 3.5 Flash-Lite, Gemini 3.1 Flash-Lite, Gemini 2.5 Pro, Gemini 2.5 Flash, Gemini 2.5 Flash-Lite, Gemini Embedding 2, several Live, Transcribe and Translate audio models, and the Gemini Robotics ER 2 previews (pricing).
• Free Google Search grounding, within limits. Gemini 3+ models share 5,000 free grounded search requests per month. Gemini 2.5 Flash and Flash-Lite share 500 free grounded requests per day on the free tier (pricing).
• Context caching on some models. Gemini 3.8 Flash, 3.6 Flash and 3 Flash Preview list context caching as free on the free tier. Flash-Lite and 2.5 models show it as "Not available" there (pricing).

These are not on the free tier: Gemini 4 Argon, Gemini 3.1 Pro Preview, the image generation models (Nano Banana 2.1, 3.1 Flash Image), Veo 3.1 and Gemini Omni Flash. All of them show "Not available" in the free column (pricing, as of October 2026).

Two catches are easy to miss. First, on unpaid services "human reviewers may read, annotate, and process your API input and output," and Google says plainly: "Do not submit sensitive, confidential, or personal information to the Unpaid Services" (terms). Second, the free tier isn't the same as the Google Cloud $300 welcome credit. That credit "can't be used towards the Gemini API or AI Studio" (billing).

A note on Pro: some third-party posts from spring 2026 say Pro was removed from the free tier in April 2026 (TinkerLLM). The official pricing page from October 19, 2026 still shows "Free of charge" for Gemini 2.5 Pro, while the newer Gemini 3.1 Pro Preview is paid. Go by the official page, and check your project's limit for 2.5 Pro in AI Studio. A model can be listed as free and still have a very low cap.

Gemini API Pricing Free Tier Limits: RPM, TPM and RPD by Model

Each Gemini model has its own RPM, TPM and RPD caps on the free tier, and Google now shows them per project in AI Studio rather than in a public table. The rate-limits doc says "Rate limits depend on a variety of factors (such as your usage tier) and can be viewed in Google AI Studio" (rate limits, updated October 15, 2026).

That's why most articles disagree with each other. A DEV Community cheatsheet from May 2026 lists 10 RPM / 250,000 TPM / 500 RPD for a 2.5 Flash preview and 30 RPM / 1,000,000 TPM / 1,500 RPD for 2.0 Flash-Lite (DEV). Those model versions are no longer the current lineup, so treat such figures as history, not as your limits.

What Google does publish officially:

And here is the per-model view the official pages support, with the cap columns left for your own project's numbers:

Prices are from the pricing page as of October 2026; check the official page before you budget. Price ranges depend on prompt length (for example, ≤200k vs >200k tokens on 3.1 Pro Preview) or input type.

To read your real caps:

1. Sign in at aistudio.google.com/rate-limit.
2. Select the project you'll call from. Limits are per project, so the default project and a second project can show different usage.
3. Write down RPM, TPM and RPD for each model you plan to use. Put them in the last column of the table above, because the cost math below runs on them.

How to Get Gemini API Free Tier Access

You get free tier access by signing in to Google AI Studio, accepting the terms and creating an API key. No credit card is needed at this stage. For new users, AI Studio creates "a default Google Cloud project and API key after you accept the Terms of Service" (API key docs).

Steps:

1. Go to the API keys page with a Google account. You must be 18 or older (terms).
2. Accept the Terms of Service. A default project and key are created for new users.
3. If you need another key, click Create API key. Since May 28, 2026, new keys are created as authorization keys tied to service accounts, which supports "fast-acting leaked key enforcement" (API key docs).
4. If you use a standard key, select Restrict to Gemini API only. The docs note that unrestricted standard keys face rejection by the Gemini API (API key docs).
5. Store the key in an environment variable, GEMINI_API_KEY or GOOGLE_API_KEY. If both are set, GOOGLE_API_KEY takes precedence (API key docs).

export GEMINI_API_KEY="your-key-here"
curl "https://generativelanguage.googleapis.com/v1beta/models/gemini-3.5-flash-lite:generateContent" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"contents":[{"parts":[{"text":"Say hello in five words."}]}]}'

The model ID in this example follows the naming on the pricing page. Confirm the exact ID in AI Studio's model picker before you run it.

Common setup mistakes and fixes:

• You already have a Google Cloud account and see no project. Existing Cloud users must import projects into AI Studio manually (API key docs). Import the project on the AI Studio Projects page.
• The key is hardcoded in a frontend app. Google's guidance is to never commit keys or hardcode them client-side (API key docs). Move calls to a server and read the key from an environment variable or secret manager.
• An old, unused key stopped working. Dormant unrestricted keys face blocking starting May 7, 2026 (API key docs). Create a new key and restrict it.

How to Use the Gemini API Free Tier Without Hitting Limits

To stay inside free tier limits, keep requests under your per-minute caps, send high-volume work to Flash-Lite, and retry 429 errors with exponential backoff. Google's troubleshooting guide recommends "exponential backoff with jitter" for 429 RESOURCE_EXHAUSTED and 503 UNAVAILABLE (troubleshooting).

A working routine:

1. Find which limit you'll hit first. For example, if your AI Studio page showed 10 RPM, 250,000 TPM and 500 RPD (sample values, not your limits), RPM allows 14,400 requests per day in theory, so the 500 RPD cap binds long before that. With 5,000 input tokens per request, TPM allows 50 requests a minute, so RPM (10) binds before TPM. Work this out once per model.
2. Throttle on your side. Put a simple rate limiter in your client set just below your RPM, for example 9 requests a minute against a 10 RPM cap, so you don't spend retries.
3. Retry only what's retryable. Retry 429, 500 and 503 with backoff. Don't retry 400, 402 or 403, because "they indicate issues like invalid API keys" (troubleshooting).
4. Shrink tokens before cutting requests. TPM counts input tokens (rate limits). Trimming a long system prompt or chat history raises how many calls fit in a minute.
5. Split jobs by model. Use Flash-Lite for classification, tagging and extraction. Save Flash or 2.5 Pro calls for the steps that need them. Each model has its own caps, so spreading work across models also spreads the load.
6. Plan around the reset. RPD resets at midnight Pacific time (rate limits). A nightly batch that starts at 23:50 Pacific will hit the reset partway through, which can help or confuse you.

A minimal backoff sketch in Python:

import random, time

def call_with_backoff(fn, max_tries=6):
    for attempt in range(max_tries):
        try:
            return fn()
        except Exception as e:
            code = getattr(e, "code", None)
            if code not in (429, 500, 503):
                raise  # don't retry 400/402/403
            time.sleep(min(60, 2 ** attempt) + random.random())
    raise RuntimeError("Gave up after retries")

Pitfalls:

• Creating extra keys to get more quota. Limits are per project, not per key (rate limits), so five keys in one project share one bucket.
• Forgetting that grounding has its own cap. On 2.5 Flash and Flash-Lite, the 500 free grounded requests per day are shared between the two models (pricing).
• Using free tier for anything private. Human review applies to unpaid services (terms). Staying under the cap doesn't change that.

Gemini API Free Tier Cost: When Do You Actually Start Paying?

You start paying only after you link a billing account to a project. That moves the project to Tier 1, and from then on your usage on that project is billed at paid rates. The free tier itself never charges you: "Users on the Free Tier incur no charges" (billing). Going over a free cap gets you a 429, not a bill.

So the real question is when the free caps start costing you more in lost throughput than paid usage would cost in money. Work it out from your own traffic.

Example A: a support-ticket tagger on Gemini 3.5 Flash-Lite (illustrative numbers). Say you run 1,000 requests a day, each with 2,000 input tokens and 500 output tokens.

• Daily tokens: 2,000,000 input and 500,000 output.
• Paid cost: 2 × $0.30 + 0.5 × $2.50 = $0.60 + $1.25 = $1.85/day, about $55.50 per 30 days (pricing, as of October 2026).
• On free tier: this only fits if your project's Flash-Lite RPD is at least 1,000. If AI Studio shows a lower RPD, the extra requests fail every day, and that's the point where paying makes sense.
• With the Batch API, which the pricing page lists at a 50% cost reduction, the same workload is roughly $27.75 per 30 days, if the work can wait.

Example B: the same workload on Gemini 3.8 Flash.

• Through December 31, 2026: 2 × $0.75 + 0.5 × $3.75 = $1.50 + $1.875 = $3.375/day, about $101 per 30 days.
• From January 1, 2027 the listed price doubles to $1.50 input and $7.50 output, giving $6.75/day, about $202.50 per 30 days (pricing).

That scheduled price change matters for budgets. A workload that sits comfortably inside the Tier 1 spend cap of $250 (billing) in December gets close to it in January. If you're building on 3.8 Flash or 3.6 Flash now, budget at the 2027 price.

A quick rule for the break-even point. Find the model's binding daily limit (usually RPD) and compare it with your expected peak daily requests. If peak is under about 70% of the cap and the data isn't sensitive, the free tier is fine. If peak goes over the cap, or you serve EEA/UK users, the free tier isn't an option anymore and you're paying either way. The 70% figure is a planning margin, not a Google number. For cross-vendor per-token costs, see our cheapest LLM API cost-per-1M-tokens table.

Pitfalls when you switch to paid:

• Thinking the free allowance carries over. Once a project is on a paid tier, its usage is billed at the paid-tier prices on the pricing page. The free tokens don't continue as a free quota inside the paid tier.
• Prepay credits running out. On Prepay, a depleted balance returns 402 Payment Required. Google says to add credits or turn on auto-reload, and not to retry (API errors). Prepay credits also expire after 12 months and are non-refundable (billing).
• Hitting the tier spend cap. When the cap is reached, "service pauses... until next billing cycle" (billing). Set a lower project-level monthly cap so a runaway loop stops early.

Gemini API Free Tier vs Paid Tier: What Changes

The paid tier raises rate limits, opens paid-only models and Batch pricing, and stops Google from using your prompts to improve its products. On paid services, "Google doesn't use your prompts... or responses to improve our products" (terms).

Sources: rate limits, billing, terms, pricing, as of October 2026; check the official pages before you commit.

Moving from Free to Tier 1 "will typically take effect instantly," and later upgrades happen "within 10 minutes" (rate limits). The Batch API also has its own limits: 100 concurrent batch requests, a 2 GB input file limit and 20 GB file storage (rate limits). Our Batch API discount comparison shows how Gemini's 50% compares with OpenAI and Claude.

To upgrade a project:

1. Open API keys or Projects in AI Studio.
2. Find the project and click Set up billing in the Billing Tier column.
3. Select or create a Google billing account and confirm payment details.
4. Buy prepay credit (minimum $5) or choose Postpay (billing).
5. Set a project-level monthly spend cap, then check Dashboard > Usage to confirm traffic is being metered.

Pitfall: the tier spend cap applies at the billing-account level across all linked projects (billing). A staging project burning tokens can pause production if both share one billing account. Give production its own billing account or a strict project cap on staging.

Gemini API Free Tier Not Working or API Key Rejected: Common Fixes

Most "not working" reports come from four things: quota 429s mistaken for a key failure, an unsupported region, a blocked or unrestricted key, or a project billing state that doesn't match what you expect. The error code tells you which one. Read it before you regenerate anything.

Error meanings from API errors and troubleshooting (as of October 2026).

Work through it in this order:

1. Read the status code. A 429 means the key works and you're over a cap. Making a new key in the same project won't help, because limits are per project.
2. Check the model. Calling Gemini 3.1 Pro Preview, Veo 3.1 or an image model from a free project fails, because those are "Not available" on the free tier (pricing).
3. Check the key's status. If you see "Your API key was reported as leaked. Please use another API key," the key has been blocked. Create a new one, and if you see unexpected charges, file a billing support case (troubleshooting).
4. Check restrictions. Unrestricted standard keys can be rejected. Choose Restrict to Gemini API only in AI Studio (API key docs).
5. Check your location. AI Studio and the Gemini API are available in a listed set of countries. Countries such as China, Iran, Syria and North Korea are not on it (available regions). Outside supported areas, Google points to the Gemini API in Gemini Enterprise Agent Platform on Google Cloud.
6. Check the environment variable. If both GOOGLE_API_KEY and GEMINI_API_KEY are set, the SDK uses GOOGLE_API_KEY (API key docs). An old value there will override the new key you just pasted.

Next steps

You can sort out your free-tier plan today with three DIY actions.

Do today:

1. Open aistudio.google.com/rate-limit, write down RPM, TPM and RPD for the two models you'll use, and work out which limit binds first, using the method in the "without hitting limits" section.
2. Estimate your daily requests and tokens per request, then run the Example A math with the official prices. For Gemini 3.8 Flash or 3.6 Flash, use the January 2027 price.
3. Add 429 backoff and a client-side rate limiter before you share the app with anyone.

If you'd rather not manage this yourself: turn on a paid Tier 1 project with a $5 prepay and a project spend cap (billing). That buys you higher limits and private prompts, and the cost is whatever you use (as of October 2026; check the official page). It suits small teams shipping to real users or handling customer data. It's overkill for weekend prototypes, and one limitation is that the Tier 1 spend cap is $250, so heavier workloads have to wait for Tier 2 ($100 paid plus 3 days). If you already run on Google Cloud and need enterprise controls, Google offers the Gemini API through Gemini Enterprise Agent Platform (available regions). It brings more setup than AI Studio, so it doesn't suit solo developers.

Related guides on Per Million:

• LLM API Pricing Comparison 2026: 5 Providers, Real Costs
• Cheapest LLM API in 2026: Real Cost per 1M Tokens Table
• Batch API Discount 2026: OpenAI vs Claude vs Gemini Table
• Claude API Pricing vs Max Plan: Break-Even Table for 2026

FAQ

What is the Gemini API free tier limit?

There's no single number. Each model has its own RPM, TPM and RPD caps, they apply per project, and RPD resets at midnight Pacific time (rate limits). Google shows your current caps in AI Studio at aistudio.google.com/rate-limit rather than in the public docs.

What is a Gemini API key free tier?

It's an API key created in Google AI Studio for a project with no billing account linked. Calls made with it use the free tier: no charges, free tokens on eligible models, and free-tier rate limits (billing). The key itself isn't "free" or "paid". The project's billing tier decides that.

Is Gemini API usage on the free tier tracked per project or per account?

Per project. Google states that rate limits are "applied per project, not per API key" (rate limits). Paid tier spend caps, on the other hand, are enforced at the billing-account level across all linked projects (billing).

What is Google Gemini API free tier compared to other LLM free tiers?

Gemini's free tier gives free tokens on production models like Gemini 3.8 Flash and 2.5 Pro without a card (pricing), but the trade-off is that Google may use and human-review your data (terms). Our 5-provider pricing comparison covers how paid rates compare once you outgrow it.

Why is my Gemini free tier API key not working?

Check the status code first. A 429 means a quota cap, not a bad key. A 403 points to permissions or key restrictions. A 400 FAILED_PRECONDITION points to billing state (API errors). Also check that the key hasn't been blocked as leaked, that it's restricted to the Gemini API, and that you're in a supported region.

Sources

• Gemini Developer API pricing (Google, updated October 19, 2026)
• Rate limits (Google, updated October 15, 2026)
• Billing (Google, updated September 28, 2026)
• Using Gemini API keys (Google, updated October 6, 2026)
• Troubleshooting guide (Google, updated October 13, 2026)
• API errors (Google, updated September 20, 2026)
• Available regions (Google, updated April 28, 2026)
• Gemini API Additional Terms of Service (Google, updated April 28, 2026)
• Google AI Studio API keys and rate limit page (sign-in required)
• Gemini API Cheatsheet 2026 (DEV Community, May 2026; unofficial, older model figures)
• Gemini API Rate Limits: Free Tier Quotas 2026 (TinkerLLM, May 2026; unofficial)

Spotted something out of date or wrong? Tell us at shigoto.ai.note@gmail.com. We check every guide against its official sources.