Gemini API Pricing (October 2026): Every Model, the January 1 Price Doubling, and What a Real Workload Costs

Gemini API pricing: the short answer

As of October 4, 2026, Google's rate card prices Gemini 3.8 Flash, its newest Flash model, at $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026. On January 1, 2027 that becomes $1.50 and $7.50. Gemini 3.1 Pro Preview is $2 and $12. Batch requests are half price. There is a free tier, but on it Google uses your content to improve its products. Priced on one real workload, Flash is less than half of Claude Sonnet 5.5 or GPT-6.1 Sol today, and still cheaper after the increase.

The headline everyone else skips is the date. Google has already published next year's price, and it is double this year's. If you are choosing a model this quarter, you are choosing at a promotional rate.

Every Gemini text model, one table

Paid tier, Standard requests, per million tokens, read from Google's pricing page today (it says it was last updated October 1):

Output prices include thinking tokens, which matters more than it looks: a model that "thinks" before answering bills that reasoning as output. Gemini 4 Argon, announced September 30, isn't on the rate card yet; its intro and list prices, and how they line up against Claude, are in Claude vs Gemini.

The January 1 doubling, and the odd result it creates

Every 3.6, 3.7 and 3.8 Flash rate on the page carries two prices: one "through December 31, 2026" and one "starting January 1, 2027." Input goes from $0.75 to $1.50, output from $3.75 to $7.50, cached input from $0.075 to $0.15, and Batch, Flex and Priority all double with it.

That produces something you don't usually see on a price list. Right now the newest Flash is cheaper than the older one: Gemini 3.5 Flash is $1.50 in and $9.00 out, so 3.8 Flash costs half as much on input and well under half on output. On January 1 the input prices match at $1.50, and 3.8 Flash is still cheaper on output ($7.50 against $9.00). If anything you run is still on 3.5 Flash, moving it now saves money twice: once today, and again next year.

Two practical points. Budget 2027 at the January price, not today's. And if you are testing Flash this quarter to decide whether to build on it, multiply your test bill by two before you compare it to anything.

What one workload costs across Google, Anthropic and OpenAI

Rate cards are hard to compare because nobody's workload is "one million tokens." So here is one shape priced everywhere: 50,000 requests a month (an internal assistant or a support-triage tool at a small company), each with about 3,000 tokens in and 600 tokens out. That's 150 million input and 30 million output tokens a month, Standard tier, no caching, each price from the vendor's own page today.

The spread is 44 to 1 from the cheapest line to the most expensive, on the same work. The full Claude and OpenAI rate cards, including caching and long-context rules, are on the Anthropic API pricing and OpenAI API pricing pages. The bill is the easy part; which of these actually does the job well enough is the question to test before you build, and it is most of what an AI implementation engagement spends its first week on.

Is the Gemini API free?

Partly. Google's own table lists a free tier with "limited access to certain models" and free input and output tokens, and says Google AI Studio usage "is free of charge in all available regions." Most Flash and Flash-Lite models have a free tier. Gemini 3.1 Pro Preview doesn't.

The real cost of free is one row in that same table: "Used to improve our products." On the free tier the answer is Yes. On the paid tier it is No. That's the line that should decide it for a business. Testing with public data on the free tier is fine. Customer emails, contracts or anything you'd mind seeing in a training set belong on the paid tier, even if the bill is a few dollars. The free tier's limits also change; Google's page says rate limits are "subject to change," so don't build a product on them. The separate question of how much you can use the Gemini app itself, on the consumer plans, is in Gemini usage limits.

The extras that add to the bill

• Batch and Flex. Half the Standard price for work that can wait. On 3.8 Flash that's $0.375 in and $1.875 out until December 31.
• Priority. 1.8 times Standard on the Flash models ($1.35 / $6.75 on 3.8 Flash) for requests that can't be slowed down.
• Grounding with Google Search. On Gemini 3 models, 5,000 search requests a month are free, shared across all of them, then $14 per 1,000. One user request can trigger several searches, and Google bills each one.
• Long prompts on Pro. Above 200,000 tokens, 3.1 Pro Preview goes from $2 / $12 to $4 / $18.
• Image and video. Priced separately. One deadline already passed: Google's page says Gemini 2.5 Flash Image was shut down on October 2, 2026, so anything still calling it needs moving to 3.1 Flash Image.

What I'd do this week

If you already call Gemini, look up which model each feature uses. Anything on 3.5 Flash moves to 3.8 Flash today. Then put January's price, not today's, in next year's budget. If you're choosing a vendor, run your real prompts through two models from the cost table above for a week and compare quality per dollar, not price per token.

Sources

• Google AI for Developers, Gemini Developer API pricing: every Gemini rate, the January 1, 2027 step, free versus paid tier, grounding, the 2.5 Flash Image shutdown (page last updated October 1, 2026; read October 4, 2026)
• Claude pricing: Opus 5.5, Sonnet 5.5 and Haiku 4.5 API rates (read October 4, 2026)
• OpenAI API docs, Pricing: GPT-6.1 Sol and GPT-6 Luna standard rates (read October 4, 2026)