Gemini API Pricing and Free Tier Limits (November 2026): Gemini 4 Argon, 3.8 Flash, Flash-Lite and Nano Banana Costs

Google’s Gemini API is still the easiest way to build with a top-tier model for free, but the details matter more than ever. The newest flagship, Gemini 4 Argon, has no free tier at all, and Google’s pricing page already lists price rises: Argon doubles on December 1, 2026, and Gemini 3.8 Flash doubles on January 1, 2027. If you are building today, you want to know both the current price and the one coming next.

Everything below comes from Google’s official Gemini API pricing and rate limit pages, checked on November 5, 2026. Prices are in US dollars per million tokens.

Gemini API Prices by Model

Gemini 4 Argon is Google’s strongest model and the one that scores 53 on the Artificial Analysis Intelligence Index (high effort), level with GPT-6 Astra and Claude Fable 5.1. Gemini 3.8 Flash is the workhorse: free to try, cheap to scale, and good enough for most apps. Note that independent coverage says Gemini 2.5 models are only available to projects that used them before, so new projects should build on 3.x or newer.

Scheduled Price Rises

Google lists two increases on its pricing page, so budget for them now:

After December 1, Gemini 4 Argon will cost the same as Claude Opus 5.5 ($4 / $20). Until then it is one of the cheapest frontier models you can buy.

What the Gemini API Free Tier Includes

You can get a free API key in Google AI Studio and start calling these models without a card:

• Text: Gemini 3.8 Flash, Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, Gemini 3.1 Flash-Lite and Gemini 3 Flash Preview.
• Open models: Gemma 4 (gemma-4-31b-it and gemma-4-26b-a4b-it), according to Apidog’s October 2026 review.
• Audio and embeddings: Gemini 3.8 Live, the TTS models, Gemini 3.5 Transcribe and Gemini Embedding 2.

Not free: Gemini 4 Argon, Gemini 3.1 Pro Preview, the Nano Banana image models (2, 2.1 and Pro) and Gemini Omni Flash video.

Three catches to know before you ship anything on a free key:

1. Your data may be used. Google’s pricing page marks free tier content as “used to improve our products”. Apidog’s summary of the terms adds that human reviewers may read, annotate and process free tier input and output. Paid tier content is not used. Never send private customer data through a free key.
2. Regional rules. In the EEA, Switzerland and the UK, only paid services may be used when you make an API client available to users, so free keys are for personal testing there.
3. No free Google Search grounding on most current models. On the free tier, grounding is either unavailable or capped.

If you are worried about privacy and cost, a local model is the other route. My best Ollama models list includes Gemma-class models that run on an 8 to 16 GB GPU.

Rate Limits and Usage Tiers

Gemini rate limits are measured in requests per minute (RPM), input tokens per minute (TPM) and requests per day (RPD). They apply per project, not per API key, so creating more keys does not give you more quota. Daily limits reset at midnight Pacific time. Google no longer prints per-model numbers in its docs; your live limits are shown in the rate limit page of Google AI Studio.

Upgrades are automatic: Free to Tier 1 is usually immediate and later upgrades apply within about 10 minutes. Going over any limit returns 429 RESOURCE_EXHAUSTED. Priority inference gets 0.3x the normal rate limit, and Grounding with Google Search on Gemini 3 and newer defaults to 1,500 requests per day per project.

Billing details that catch people out (from Apidog’s October 2026 review): paid usage requires a prepaid balance of at least $5, Gemini API usage no longer counts against the $300 Google Cloud free trial credit, unused prepaid credit expires after 12 months, and when the balance hits $0 every key on that billing account returns HTTP 402 until you top up.

Nano Banana Images, Live Audio and TTS

Interesting detail: the newer Nano Banana 2.1 is roughly half the price per image of Nano Banana 2. Batch halves image output again.

Context Caching and Grounding

• Context caching: Gemini 4 Argon cached input is $0.10 per MTok (rising to $0.20 on December 1) and is implicit only. Gemini 3.8 Flash cached input is $0.075 plus $0.50 per MTok per hour of storage (both double on January 1). Gemini 3.1 Pro caching is $0.20 (or $0.40 over 200K) plus $4.50 per MTok per hour.
• Grounding with Google Search or Google Maps: 5,000 grounded prompts free per month on paid tiers, then $14 per 1,000.

Cost Examples

Same support bot workload I use in my other pricing guides: 90M input and 12M output tokens a month.

For images, 1,000 standard 1K pictures cost about $33.60 with Nano Banana 2.1, $67 with Nano Banana 2 and $134 with Nano Banana Pro. All figures are estimates from list prices.

How to Get a Free Gemini API Key (Step by Step)

1. Go to aistudio.google.com and sign in with a Google account.
2. Click Get API key, then Create API key. AI Studio creates a Google Cloud project for you if you do not have one.
3. Copy the key and save it as an environment variable called GEMINI_API_KEY. Do not paste it into front-end code or a public repo.
4. Try a prompt in the AI Studio playground first to confirm the model you want is on the free tier for your account.
5. When you are ready for production, open the same project in AI Studio and choose Set up billing to link a Cloud Billing account. That moves the project to the paid tier, raises your limits and stops Google using your prompts to improve its products.

Quickstart: Your First Gemini API Call

With Python and Google’s official google-genai SDK:

pip install google-genai

from google import genai

client = genai.Client()  # reads GEMINI_API_KEY from the environment
resp = client.models.generate_content(
    model="gemini-3.8-flash",
    contents="Explain context caching in two sentences.",
)
print(resp.text)
print(resp.usage_metadata)  # token counts you are billed for

Or with curl:

curl "https://generativelanguage.googleapis.com/v1beta/models/gemini-3.8-flash:generateContent" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"contents":[{"parts":[{"text":"Say hello in five words."}]}]}'

Check usage_metadata (or usageMetadata in the raw JSON) on every response. It shows prompt, cached and output tokens, including the “thinking” tokens that are billed as output.

Gemini vs OpenAI vs Claude vs Grok vs DeepSeek Prices

Gemini’s real advantage is the free tier for prototyping and the very low Flash-Lite prices. For a full side-by-side with caching and batch, see my AI API pricing comparison and OpenAI API pricing guide.

Common Gemini API Errors and Fixes

• 429 RESOURCE_EXHAUSTED: you hit RPM, TPM or the daily request cap for your project. Wait (daily caps reset at midnight Pacific), add exponential backoff, or enable billing. Extra API keys in the same project do not help, because limits are per project.
• 400 FAILED_PRECONDITION / free tier not available in your country: the free tier is not offered in your region. Enable billing on the project.
• 403 PERMISSION_DENIED: the key is wrong, restricted to another API, or the model (for example Gemini 4 Argon) needs a paid project.
• 404 NOT_FOUND for a model: the model ID is misspelled or retired. Gemini 2.5 models are only available to projects that used them before.
• 500 INTERNAL / 503 UNAVAILABLE: temporary overload on Google’s side. Retry with backoff, cut very long prompts, or switch to a Flash model for a while.
• Bill higher than expected: look at thinking tokens (lower the thinking level for simple tasks), Search grounding beyond the free 5,000 prompts, and long prompts over 200K on Pro models, which cost more per token.

Frequently Asked Questions

Is the Gemini API free?

Partly. Gemini 3.8 Flash, 3.6 Flash, the Flash-Lite models, Gemma 4 and several audio models have a free tier with rate limits. Gemini 4 Argon, 3.1 Pro and the image models are paid only.

How much does Gemini 4 Argon cost?

$2 input and $10 output per million tokens until November 30, 2026, then $4 and $20 from December 1.

What are the Gemini free tier limits?

Limits are per project in RPM, TPM and RPD, reset daily at midnight Pacific, and are shown in Google AI Studio for your account. Google no longer publishes fixed per-model numbers.

Does Google train on my Gemini API data?

On the free tier, Google says content is used to improve its products. On the paid tier it is not.

Is Gemini cheaper than ChatGPT or Claude?

Flash-Lite and Flash are among the cheapest capable models anywhere. At the top end, Argon is cheaper than GPT-6 Astra and, until December 1, cheaper than Claude Opus 5.5. See the full AI API pricing comparison.

Do I need a credit card for the Gemini API?

No. A Google account and an AI Studio key are enough for the free tier. You only add billing for paid models, higher limits or production use.

Is Google AI Pro the same as the Gemini API?

No. Google AI Pro is the consumer subscription for the Gemini app. The API is billed separately per token through Google Cloud.

Will Gemini 4 Argon really double in price?

Google’s pricing page lists $4 input and $20 output from December 1, 2026, up from $2 and $10. Budget for the new price if you are launching now.

How do I use Gemini with the OpenAI SDK?

Google offers an OpenAI-compatible endpoint, so many OpenAI-style libraries work by changing the base URL and key. The native google-genai SDK exposes more Gemini features, such as caching and grounding.

Final Thoughts

The Gemini API remains the best place to prototype for free, and Flash-Lite is one of the cheapest production models available. Just plan for the two price rises: Gemini 4 Argon doubles on December 1 and Gemini 3.8 Flash doubles on January 1, 2027. Keep private data off free keys, and move to a paid tier before you launch.

Comparing providers? Read my Claude API pricing, OpenAI API pricing and DeepSeek API pricing guides.

Sources

• Google: Gemini API pricing (official)
• Google: Gemini API rate limits (official)
• Apidog: Gemini API free tier in 2026 (Oct 23, 2026)
• Artificial Analysis Intelligence Index leaderboard snapshot