Gemini API Pricing 2026: Rates, Free Tier & Cost Calculator
Gemini is the only major LLM API you can build on for $0 — and the only one where making your prompt slightly longer can more than double the bill. Both of those facts are on this page, with the numbers.
Prices verified July 28, 2026 against ai.google.dev/gemini-api/docs/pricing.
Start here: Gemini has a real free tier, and a real catch
Google's pricing page lists "Free of charge" input and output on several models — including Gemini 3.6 Flash, 3.5 Flash, 2.5 Flash and 2.5 Flash-Lite — subject to free-tier rate limits. This is not trial credit that expires. It is a standing allowance, and no other provider in this comparison offers one.
The catch is printed on the same page. On the free tier, "Content used to improve our products" is marked Yes. Your prompts and the model's responses feed Google's product improvement. That is a fine trade for a side project or a spike. It is not a fine trade for customer data, proprietary source code, or anything with a compliance obligation attached — move to the paid tier before that traffic starts flowing, not after.
Gemini API Cost Calculator
Estimate a paid-tier Gemini bill, and add models from OpenAI, Anthropic, xAI or DeepSeek to compare on the identical workload. Runs entirely in your browser.
Models to compare (5 selected)
| Model | Input cost | Output cost | Per request | Monthly total |
|---|---|---|---|---|
| Gemini 2.5 Flash-LiteGoogleCHEAPEST | $14.76 | $4.00 | $0.00094 | $18.76 |
| Gemini 3.5 Flash-LiteGoogle | $44.28 | $25.00 | $0.0035 | $69.28 |
| Gemini 2.5 FlashGoogle | $44.28 | $25.00 | $0.0035 | $69.28 |
| Gemini 2.5 ProGoogle | $184.50 | $100.00 | $0.01 | $284.50 |
| Gemini 3.6 FlashGoogle | $221.40 | $75.00 | $0.01 | $296.40 |
The calculator prices the Standard synchronous tier. Below 200k-token prompts. Batch mode halves these figures; Priority raises them. See the service-tier section below.
Your token budget is settled. Your AI discoverability probably isn't.
Gemini and ChatGPT are both answering “which tool should I use?” for your category right now. Run a free scan to see whether your product shows up in those answers, and where it ranks when it does.
On this page
The 200k context cliff: where Gemini bills stop being linear
Every other provider in this comparison charges one input rate regardless of prompt length. Google charges two. On Gemini 2.5 Pro the rate is $1.25 per million input tokens for prompts up to 200,000 tokens and $2.50 above it — and critically, the higher rate applies to the entire prompt, not just the portion past the line. Gemini 3.1 Pro carries the same structure.
Input cost of a single Gemini 2.5 Pro request as the prompt grows:
| Prompt size | Rate applied | Input cost per request | Band |
|---|---|---|---|
| 150,000 tokens | $1.25 / 1M | $0.19 | Standard band |
| 190,000 tokens | $1.25 / 1M | $0.24 | Standard band |
| 210,000 tokens | $2.50 / 1M | $0.53 | Above 200k — doubled rate |
| 400,000 tokens | $2.50 / 1M | $1.00 | Above 200k — doubled rate |
Look at rows two and three. Going from 190,000 to 210,000 tokens — an 11% larger prompt — takes the input cost from $0.24 to $0.53, a 121% jump. Adding two more retrieved documents to a RAG prompt, or one more turn of history to a long conversation, can cost you more than doubling your traffic would.
What to do about it, in order of how much it saves:
- Instrument prompt length, not just cost. Log a histogram of input token counts per request. A bill that grew 40% for no obvious reason is usually a distribution that crept across 200k, and you cannot see that in an average.
- Set a hard truncation just under the line. If your prompts cluster around 200k, capping retrieval at roughly 190k tokens is free money — you lose a couple of low-ranked chunks and halve the input rate.
- Split, do not stretch. Two 150k-token requests cost $0.38 in input at the standard band. One 300k-token request costs $0.75. Same tokens, double the price, for choosing one call over two.
- Check whether you need Pro at all. Gemini 2.5 Flash is a flat $0.30 at any length. On genuinely long-context work that does not need Pro-level reasoning, Flash both avoids the cliff and starts 4x cheaper.
Gemini API price table, per 1M tokens
Paid Standard tier, USD, text tokens, prompts under 200,000 tokens. Audio, image and video input are tokenised at separate rates not shown here.
| Model | Input / 1M | Cached input / 1M | Output / 1M | Batch input / 1M | Notes |
|---|---|---|---|---|---|
Gemini 3.6 Flashgemini-3.6-flash | $1.50 | $0.15 | $7.50 | $0.75 | Flat rate at any prompt length |
Gemini 3.5 Flashgemini-3.5-flash | $1.50 | $0.15 | $9.00 | $0.75 | Flat rate at any prompt length |
Gemini 3.5 Flash-Litegemini-3.5-flash-lite | $0.30 | $0.03 | $2.50 | $0.15 | Flat rate at any prompt length |
Gemini 3.1 Pro (preview)gemini-3.1-pro | $2.00 | $0.20 | $12.00 | $1.00 | Rate shown is for prompts ≤200k tokens; doubles above that |
Gemini 2.5 Progemini-2.5-pro | $1.25 | $0.13 | $10.00 | $0.63 | ≤200k-token prompts; $2.50 / $15 above |
Gemini 2.5 Flashgemini-2.5-flash | $0.30 | $0.03 | $2.50 | $0.15 | Flat rate at any prompt length |
Gemini 2.5 Flash-Litegemini-2.5-flash-lite | $0.10 | $0.01 | $0.40 | $0.05 | Cheapest Gemini |
Gemini 2.0 Flashgemini-2.0-flash | $0.10 | $0.03 | $0.40 | $0.05 | Flat rate at any prompt length |
Track Gemini repricing without re-reading the docs
Google ships new Flash and Pro tiers often, and they usually land under the ones they replace. We watch the pricing pages and email you when a figure above changes.
Already on the calendar: Claude Sonnet 5's introductory $2 / $10 per 1M tokens expires Aug 31, 2026 and goes to $3 / $15 — a 50% increase on the same workload. Prices here were last re-verified July 28, 2026.
Batch, Standard and Priority: paying for latency
Google is the only provider here that sells the same model at three different prices according to how fast you need it. Everyone offers a batch discount; only Google offers a paid upgrade.
Batch — 50% off
Asynchronous, results back within a day. Gemini 3.6 Flash input drops from $1.50 to $0.75. Use it for anything with no human waiting.
Standard — baseline
The synchronous rate, and what every number on this page shows unless stated otherwise.
Priority — premium
Lower, more predictable latency for a price uplift. Gemini 3.6 Flash input is listed at $2.70 against $1.50 Standard — an 80% premium.
The useful consequence is that latency becomes a budget line you can explicitly trade, rather than something you hope for. Most products have both shapes of traffic inside them: an interactive path where a slow response is a visible defect, and a background path — enrichment, indexing, summarising yesterday's data — where it is invisible. Routing the second to Batch at half price funds Priority on the first, and the blended rate can land below what you were paying on Standard for everything.
Context caching: cheap tokens, but you rent the shelf
Cached input on Gemini costs about 10% of the standard rate, in line with OpenAI and Anthropic. What is different — and what nobody else on this comparison charges — is that Google bills $1.00 per 1,000,000 tokens per hour in storage for as long as the cache exists, whether or not anything reads it.
| Cached corpus | Storage / hour | Storage / day | Storage / 30 days |
|---|---|---|---|
| 50,000 tokens | $0.05 | $1.20 | $36.00 |
| 200,000 tokens | $0.20 | $4.80 | $144.00 |
| 1,000,000 tokens | $1.00 | $24.00 | $720.00 |
A one-million-token reference corpus held permanently in cache costs $720.00 a month before a single request touches it. That is fine if the corpus serves a hundred thousand requests. It is pure waste if it serves two hundred.
The practical rule: Gemini caching is a throughput optimisation, not a latency one. Cache large prefixes that are hit continuously during business hours, and let the cache lapse overnight rather than paying rent on an idle shelf. Compare this to Anthropic, which charges a one-time write premium but nothing for residency, and OpenAI, which charges neither — the right caching strategy is genuinely provider-specific, and copying your OpenAI approach across to Gemini will quietly cost you money.
Flash-Lite, Flash or Pro?
Flash-Lite — $0.10 / $0.40
Mechanical, high-volume work: classification, tagging, routing, extraction into a fixed schema, moderation pre-filters. At these rates the model cost of a million classifications is smaller than the engineering time spent arguing about it. Flat rate at any prompt length — no cliff.
Flash — $0.30 / $2.50
The default. Handles chat, summarisation, RAG answering and most content work. Also flat-rate at any length, which makes it the model of choice for genuinely long-context jobs where Pro's doubling would bite. Free-tier eligible.
Pro — $1.25 / $10.00 under 200k, $2.50 / $15.00 above
Reasoning, code, multi-step analysis, synthesis across many documents. Worth the money when the task actually fails on Flash — and only then, because Pro is 4x Flash on input before the cliff and 8x after it.
AI Studio and Vertex AI are billed separately
Google sells Gemini through two doors and it matters for your invoice. The Gemini Developer API (via Google AI Studio) is what this page prices; it is where the free tier lives and it bills to an API key. Vertex AI is the Google Cloud path, billed through a GCP project, with IAM, VPC controls, regional deployment and committed-use options — and no equivalent standing free tier.
Prototypes almost always start on AI Studio because it is free and takes a minute to set up. Production systems with compliance requirements usually end up on Vertex. Budget the migration deliberately: the models are the same, but the pricing structure, the data-handling terms and the free allowance are not.
Gemini API pricing FAQ
Is the Gemini API actually free?
Genuinely yes, within limits, and Gemini is the only API in this comparison with a standing free tier rather than one-off promotional credit. Google's pricing page lists "Free of charge" input and output on several models including Gemini 3.6 Flash, 3.5 Flash, 2.5 Flash and 2.5 Flash-Lite, subject to free-tier rate limits. The catch is stated on the same page: on the free tier, "Content used to improve our products" is marked Yes. Your prompts and responses feed Google's product improvement. For a hobby project or a prototype that is usually fine; for anything carrying customer data, confidential source code or regulated content, move to the paid tier before you ship, not after.
Why did my Gemini bill double when I made prompts longer?
Because you crossed 200,000 tokens in a single prompt. Google prices several Gemini models in two bands: Gemini 2.5 Pro is $1.25 input / $10.00 output for prompts up to 200k tokens, and $2.50 / $15.00 for prompts above it — and the higher rate applies to the whole prompt, not just the tokens past the threshold. A request at 210,000 tokens costs more than twice a request at 190,000. This is the single most expensive surprise in Gemini pricing, and it is invisible until you look at a per-request cost breakdown.
What is the difference between Gemini Flash, Flash-Lite and Pro?
They are three price tiers, and the gap between them is much wider than the naming suggests. On the 2.5 generation, Flash-Lite is $0.10 / $0.40, Flash is $0.30 / $2.50, and Pro is $1.25 / $10.00 below the 200k threshold. That makes Pro 12.5x Flash-Lite on input and 25x on output. Flash-Lite exists for high-volume mechanical work — classification, tagging, extraction, routing. Flash is the general default. Pro is for reasoning, long-document synthesis and code, and it is the only one of the three that hits the 200k price cliff.
Does Gemini context caching cost anything beyond the reduced token rate?
Yes, and this is where Google differs from OpenAI and Anthropic. Cached input tokens bill at roughly 10% of the standard rate, the same as elsewhere — but Google additionally charges storage at $1.00 per 1,000,000 tokens per hour for as long as the cache exists. Nobody else on this comparison bills for cache residency by time. A 100,000-token cached corpus therefore costs $0.10 per hour, or about $72 a month, whether or not a single request touches it. Caching wins decisively when a large prefix is hit constantly; it can lose money outright when a big cache sits idle overnight.
What are Gemini's Batch and Priority service tiers?
Google sells the same model at different prices depending on how urgently you need the answer. Batch mode takes 50% off — Gemini 3.6 Flash drops from $1.50 to $0.75 per million input tokens — in exchange for asynchronous delivery. Priority goes the other way, charging a premium for lower and more predictable latency: the same Gemini 3.6 Flash is listed at $2.70 per million input on Priority, an 80% uplift over Standard. No other provider in this comparison sells a paid latency upgrade. If you have a user-facing path where p99 latency is a product requirement and a background path where it is not, Gemini is the only API where you can price both explicitly.
Is Gemini cheaper than GPT or Claude?
At the bottom of each range, Gemini is competitive rather than dominant: Gemini 2.5 Flash-Lite at $0.10 / $0.40 is line-for-line identical to GPT-4.1 nano, and slightly more expensive than GPT-5 nano on input. Where Gemini genuinely wins is the free tier — no competitor lets you build and test at zero cost indefinitely — and Batch pricing on Flash models for bulk work. Where it loses is long-context Pro workloads, because the 200k doubling can put Gemini 2.5 Pro above GPT-5 on exactly the long-document tasks it is otherwise good at.
Does the Gemini API bill differently through Google AI Studio and Vertex AI?
They are separate front doors to the same models with separate billing. The rates quoted on this page are the Gemini Developer API rates from Google AI Studio, which is where the free tier lives. Vertex AI is the Google Cloud path, billed through your GCP project with enterprise controls, committed-use options and regional deployment — and it does not carry the same free tier. Which one you are on determines both your bill and your data-handling terms, so confirm before comparing a Gemini quote against anything else.
Are Gemini's thinking tokens billed as output?
Yes. Gemini's reasoning models bill internal thinking at the output rate even though the thinking is not returned to you in full. On Gemini 2.5 Pro at $10.00 per million output tokens, a request whose visible answer is 400 tokens but which generated 6,000 thinking tokens costs sixteen times what the visible answer implies. When you estimate a Gemini bill, set the output figure to the total generated tokens, not the answer length.
Does audio or image input cost the same as text on the Gemini API?
No. The rates on this page are text tokens only. Several Gemini models publish a separate, higher rate for audio input, and image, video and document input are tokenised at their own rates that can differ substantially from text. A multimodal app cannot be budgeted from a text price table — check the per-modality rows on Google's pricing page for the specific model you are calling.
Per-token pricing for the other providers
Every page in this set reads from the same verified price table, so the numbers never disagree with each other.
OpenAI API pricing
GPT-5.6, GPT-5, mini and nano rates, plus the cost calculator.
Claude API pricing
Anthropic's tier ladder, cache-write premiums and the Sonnet 5 intro rate.
Grok API pricing
xAI's unusually flat input/output gap and what it means for output-heavy apps.
DeepSeek API pricing
The cheapest rates on the board and a 50x cache-hit discount.
LLM API pricing comparison
All five providers head to head, with a recommendation per workload shape.
More on Google AI
ChatGPT already recommends your Gemini-powered app. Does it recommend yours?
If you're building in LLM APIs, run a free AI-visibility scan on your own product — we ask ChatGPT across 5 prompt angles and score how often you get named. ~30 seconds, no signup, no card.
📬 Get the best new AI tools delivered weekly
One concise email with fresh launches, trending picks, and featured standouts.