Google Gemini API Pricing: Every Model, Rate & Free Tier Explained

If you are building with Google’s AI models in 2026, the first question is always the same. This Google Gemini API pricing 2026 guide answers it: what does the Gemini API actually cost? The answer has a few moving parts — a genuinely useful free tier, per-token pricing that changes on January 1, 2027, and add-ons like context caching and search grounding that can quietly reshape your bill.

This guide breaks down every current Gemini API rate, explains the free tier, walks through the scheduled 2027 price change, and shows you how to estimate (and reduce) your monthly spend. All prices below are in USD per one million tokens unless stated otherwise.

Is the Gemini API Free?

Yes — partially. Google offers a free tier for the Gemini Developer API with modest rate limits and quotas, which is enough for prototyping, learning, and low-volume side projects. You can also experiment at no cost inside Google AI Studio, which has its own free plan.

There are two important caveats to the free tier:

1. Limited model access. Not every model is available on the free tier, and quotas are restrictive compared to paid usage.
2. Data usage terms. On the free tier, your prompts and responses may be used to improve Google’s products. On the paid tier, they are not.

The moment you need production reliability, higher rate limits, or the newest models at scale, you move to the paid tier, which requires a billing account. Paid API usage is billed purely on consumption — there is no monthly subscription fee for the API itself. (Google’s consumer AI plans, AI Pro and AI Ultra, are separate products that raise limits inside AI Studio and the Gemini app; API billing is metered independently.)

For the authoritative, always-current numbers, Google’s official pricing documentation is the source of truth: Gemini API pricing.

Google Gemini API Pricing 2026 at a Glance

Here are the headline rates for the current flagship model family, valid through December 31, 2026:

Model	Input (per 1M tokens)	Output (per 1M tokens)	Notes
Gemini 3.8 Flash (Standard)	$0.75	$3.75	Date-bound pricing; doubles Jan 1, 2027
Gemini 3.7 Flash (Standard)	$0.75	$3.75	Same introductory schedule
Gemini 3.5 Flash	$1.50	$9.00	Standard list pricing
Context caching (input)	$0.075	—	Per 1M cached tokens
Context cache storage	$0.50/hr	—	Per 1M tokens stored
Search grounding	$14 per 1,000 requests	—	5,000 requests/month free

Output pricing includes thinking tokens for models that expose reasoning. Rates differ across the Batch, Flex, and Priority service tiers, which we cover below.

Gemini 3.8 Flash Pricing: The Current Standard

Gemini 3.8 Flash is the model most developers reach for in 2026: fast, capable, and aggressively priced. Google lists its standard-tier pricing as $0.75 per million input tokens and $3.75 per million output tokens — but this is explicitly date-bound pricing, not a permanent list price.

What does “date-bound” mean in practice? Google’s pricing page states these rates apply through December 31, 2026. There is no promotional language on the page — no “introductory offer” banner — just the dates. Functionally, though, it works like a launch discount: you get roughly half off for the rest of the year.

To put that in concrete terms: one million input tokens is roughly 750,000 words — about ten full-length novels. At $0.75, feeding the model ten novels’ worth of text costs less than a cup of coffee. Output is priced higher because generation is computationally more expensive, which is the industry-standard pattern.

The January 2027 Price Change You Need to Know About

This is the single most important budgeting detail in this guide. On January 1, 2027, the scheduled rates take effect and the current prices double across every line item:

Line item	Through Dec 31, 2026	From Jan 1, 2027
Standard input	$0.75	$1.50
Standard output	$3.75	$7.50
Context caching input	$0.075	$0.15
Cache storage	$0.50/hr	$1.00/hr
Batch input / output	$0.375 / $1.875	$0.75 / $3.75
Priority input / output	$1.35 / $6.75	$2.70 / $13.50

If you are planning a 2026 build with a 2027 runway, budget against the January rates, not today’s. A workload that costs $400/month in November 2026 will cost roughly $800/month in January 2027 at identical usage. Lock in architectural decisions — like aggressive context caching — now, while the cheaper rates give you room to experiment.

Context Caching: The Hidden Cost Saver

If your application sends the same large context repeatedly — a system prompt, a knowledge base, a long document — context caching is the closest thing to free money in the Gemini API.

Instead of paying full input price every time you resend that context, you pay:

• $0.075 per million cached input tokens (one-tenth of standard input pricing), plus
• $0.50 per million tokens per hour of cache storage

Cached content that gets reused across many requests can cut input costs by 80–90% for retrieval-augmented generation (RAG) workloads, chatbots with long system prompts, and document Q&A. The trade-off is the hourly storage fee, so caching only wins when you reuse the cached content enough times to offset it. As a rule of thumb: if you send the same 100K-token context more than a handful of times per hour, caching pays for itself many times over.

Batch, Flex & Priority Tiers: Pay Less for Flexible Workloads

Not every workload needs an instant response. Google offers service tiers that trade latency flexibility for lower prices:

• Batch tier — roughly 50% off standard rates ($0.375 input / $1.875 output through 2026). Designed for non-urgent, large-scale processing like bulk classification, dataset labeling, or overnight evaluations.
• Flex tier — discounted pricing for workloads that can tolerate variable capacity.
• Priority tier — premium pricing ($1.35 input / $6.75 output through 2026) for guaranteed throughput when you cannot afford throttling.

For most side projects and MVPs, Standard tier is fine. The moment you have a pipeline that processes thousands of documents and nobody is waiting on the result, Batch tier is the easiest cost win available.

Search Grounding and Other Add-On Costs

Grounding with Google Search lets the model pull in fresh, real-world information — essential for anything where the training cutoff is not enough. Pricing is request-based rather than token-based:

• 5,000 requests per month free (shared across Gemini 3.x models)
• $14 per 1,000 requests beyond that

Other cost factors to keep on your radar: audio input/output is priced separately from text (audio output runs dramatically higher per million tokens), image generation has its own rate card, and the Live API models carry separate streaming-oriented pricing. If your app is multimodal, price each modality separately instead of assuming the text rates apply.

How Gemini API Pricing Compares

Token pricing only makes sense in context. Here is how the Gemini API stacks up against the alternatives we have covered in depth:

• DeepSeek’s API has built a reputation on aggressive pricing for reasoning-heavy workloads — see our full breakdown of DeepSeek API pricing in 2026 to compare cost per reasoning task.
• If you are choosing between Google’s models and an AI coding assistant instead, our GitHub Copilot pricing guide for 2026 covers its newer usage-based billing model.
• For the search-assistant side of the market, Perplexity’s pricing shows what a bundled subscription costs versus pay-per-token API usage.

The short version: Gemini 3.8 Flash at $0.75/$3.75 is among the cheapest frontier-class options available in 2026 — but only until the January 2027 reset.

How to Estimate Your Monthly Gemini API Bill

Let us walk through a realistic example. Suppose you run a customer-support chatbot:

• 50,000 conversations per month
• Average 2,000 input tokens per conversation (with a cached system prompt)
• Average 300 output tokens per conversation

Without caching: 50,000 × 2,000 = 100M input tokens × $0.75 = $75, plus 50,000 × 300 = 15M output tokens × $3.75 = $56.25. Total: ~$131/month.

With context caching (assuming 1,500 of the 2,000 input tokens are cached): cached input = 75M tokens × $0.075 = $5.63; uncached = 25M × $0.75 = $18.75; output unchanged at $56.25; plus cache storage (say 1.5M tokens × $0.50 × 730 hours = steep — in practice you would scope cache TTLs aggressively). Even with conservative storage assumptions, caching typically cuts this bill by half or more.

The formula is always: (input tokens × input rate) + (output tokens × output rate) + add-ons. Google’s billing console shows per-model usage, so reconcile your estimates against real data after the first week.

Tips to Cut Your Gemini API Costs

1. Cache aggressively. Any repeated context over ~1,000 tokens is a caching candidate.
2. Use Batch tier for anything that does not need a real-time response.
3. Pick the smallest capable model. Flash handles the vast majority of tasks; reserve Pro-class models for the hard 10%.
4. Trim thinking tokens. If the API exposes a thinking budget, cap it — reasoning tokens bill at output rates.
5. Deduplicate requests. Cache frequent identical queries at your application layer before they ever reach the API.
6. Watch the calendar. Re-forecast every budget against January 2027 rates now, not in December.

Frequently Asked Questions

Is the Gemini API really free?

There is a free tier with usage quotas, suitable for prototyping and learning. Production workloads need the paid tier, which bills per token. Free-tier usage may be used to improve Google’s products; paid-tier usage is not.

Managing keys and fallbacks across providers is easier with a gateway — see how Portkey’s AI gateway routes and monitors API calls. how Portkey's AI gateway manages API calls

What happens to Gemini API pricing on January 1, 2027?

Current date-bound rates double. Standard input goes from $0.75 to $1.50 per million tokens and output from $3.75 to $7.50, with proportional increases across caching, Batch, Flex, and Priority tiers.

How much is 1 million tokens in plain English?

Roughly 750,000 words — about ten novels. Most chatbots use a few thousand tokens per conversation, so a million tokens stretches further than it sounds.

Do I need a Google AI Pro or Ultra subscription to use the API?

No. API billing is separate and purely usage-based. The consumer subscription plans raise limits in AI Studio and the Gemini app; the API needs only a Google Cloud billing account.

What is context caching and is it worth it?

It stores reusable prompt context so you pay $0.075 instead of $0.75 per million input tokens on repeats, plus a small hourly storage fee. For RAG and long-system-prompt apps, it is usually the biggest available saving.

Where can I see the official current prices?

Google’s own documentation: Gemini API pricing. Always confirm there before budgeting, since rates and model availability change.

Final Thoughts

Google Gemini API pricing in 2026 is genuinely developer-friendly — a usable free tier, $0.75/$3.75 flagship rates, and powerful cost levers in caching and Batch tier. The catch is the calendar: every rate on the page doubles on January 1, 2027. Build with today’s prices, budget with tomorrow’s, cache everything you repeat, and check the official pricing page before you commit to a number. Do that, and the Gemini API remains one of the best value-for-money ways to put frontier AI into production.

Gemini Developer API vs Vertex AI: Which Billing Path?

One source of confusion deserves its own section. Google sells Gemini model access through two doors:

• Gemini Developer API (ai.google.dev) — the straightforward, pay-per-token path covered in this guide. Simplest billing, fastest to start.
• Vertex AI (Google Cloud) — the enterprise path, with the same underlying models but billed through Google Cloud’s enterprise agreements, with VPC controls, committed-use discounts, and regional endpoints.

Token rates are broadly comparable, but Vertex AI adds enterprise features (and enterprise complexity). If you are an individual developer or small team, the Developer API is almost always the right door. If your company already negotiates Google Cloud commits, talk to your account team — committed-use discounts on Vertex can beat list API pricing at high volume.

Rate Limits, Quotas & What “Free” Really Gets You

Pricing is only half the story; the other half is quotas. The free tier caps you on requests per minute, requests per day, and tokens per minute — and the caps differ per model. Paid tier raises these substantially, and you can request further increases.

Practical implications:

• A free-tier prototype that works beautifully for one user can start throwing 429 errors the day ten users show up. Load-test on the paid tier before launch.
• Quotas reset on fixed windows, not rolling ones — a burst at 11:59 PM and another at 12:01 AM counts against two windows.
• If you hit limits consistently, the fix is rarely “optimize prompts” — it is “enable billing.” The paid tier’s quotas are an order of magnitude more generous.

Monitor all of this in the Google AI Studio usage dashboard and the Cloud billing console from day one, not the day the invoice surprises you.

Which Gemini Model Should You Actually Use?

Price means nothing without fit. A quick decision framework for 2026:

Your need	Best pick	Why
General chat, agents, most apps	Gemini 3.8 Flash	Best price/performance in the lineup
Cheapest possible at scale	Gemini 3.8 Flash + Batch tier	~$0.375/$1.875 through 2026
Hardest reasoning tasks	Gemini 3.x Pro-class	Higher rates, but fewer retries needed
Learning / prototypes	Free tier	Zero cost within quotas
Real-time voice	Live API models	Separate streaming rate card

The common mistake is defaulting to the most capable model “just in case.” Start with Flash, measure quality on your actual evals, and only upgrade the specific calls that fail. Most teams find Flash handles 80–90% of their traffic — and at $0.75 per million input tokens, that 80% costs almost nothing.

Frequently Asked Questions (Continued)

Does the free tier include Gemini 3.8 Flash?

Free-tier model availability is limited and changes over time — typically smaller or previous-generation models have the most generous free quotas. Check the official model documentation for current per-model availability.

Can I set spending limits on the Gemini API?

Yes — through Google Cloud billing budgets and alerts. Set a budget, attach alert thresholds (50%, 90%, 100%), and consider programmatic caps in your own backend for defense in depth. There is no hard “stop at $X” switch on the Developer API itself, so alerts plus application-level throttling is the standard pattern.

Is my API data used for training?

On the paid tier, no — your inputs and outputs are not used to improve Google’s products. On the free tier, they may be. If you handle sensitive data, use the paid tier.

How does Gemini API pricing compare to DeepSeek and OpenAI?

As of late 2026, Gemini 3.8 Flash undercuts most frontier-class competitors on list price — our DeepSeek API pricing breakdown has the side-by-side numbers. The gap narrows after the January 2027 reset, which is exactly why you should model both.