Gemini API Pricing 2026: Free Tier and Per-Token Rates

The Gemini API costs $2 per million input tokens and $10 per million output tokens on Gemini 4 Argon through 30 November 2026, rising to $4 and $20 on 1 December. Gemini 3.8 Flash costs $0.75 and $3.75 through 31 December, then doubles. The free tier covers the Flash and Flash-Lite models, but not Gemini 4 Argon or Gemini 3.1 Pro Preview.

Every price here was read on Google's Gemini API pricing page on 19 October 2026 (last updated by Google on 16 October), and every figure is in US dollars per million tokens on the Standard tier unless stated. Our take: both flagship rates are introductory, so a cost model built on today's numbers is wrong by January. Budget at the post-promotion price and treat the current one as a discount.

Output prices include thinking tokens, which matters for reasoning models: a long chain of thought bills at the output rate even though you never see it in the response.

Toolradar data: Google Gemini is listed as an alternative to 128 other tools on Toolradar, the software directory we run, fourth of 995 AI agent tools, as of October 2026. Toolradar visitors most often compare it with Microsoft Copilot, Grok and ChatGPT; for API buyers, the comparison with Claude below is the one that sets the budget.

What does the Gemini API free tier include?

The free tier gives you free input and output tokens on the Flash family, including Gemini 3.8 Flash, 3.6 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite, 2.5 Flash, 2.5 Flash-Lite and 2.5 Pro. Gemini 4 Argon, Gemini 3.1 Pro Preview and the Omni Flash models are paid only. Google AI Studio itself is free to use in every region where it is available.

The trade-off is your data. Google's pricing table marks free-tier content as used to improve its products, and paid-tier content as not used. Do not send customer data, proprietary code or anything under an NDA through a free-tier key; that rule alone moves most production workloads to the paid tier, whatever the volume.

The free tier also skips several paid features. Context caching is paid-only on several models, including 2.5 Flash and the Flash-Lite models, the Batch API is a paid-tier feature, and grounding with Google Search is listed as not available on the free tier for Gemini 3.8 Flash. The pricing page does not print free-tier rate limits and notes that rate limits are subject to change, so treat the free tier as a prototyping allowance rather than capacity you can plan around.

Which Gemini model gives the best price per task?

Gemini 4 Argon is Google's flagship reasoning model, pitched at complex software engineering, autonomous multi-step agents and long generations up to 1 million tokens. At its introductory rate it costs about 2.7 times Gemini 3.8 Flash on both input and output, and the gap holds after both promotions end.

Gemini 3.8 Flash is the default for most developer workloads. Google describes it as its most intelligent Flash model, built for long-horizon software engineering and agents, and it is the newest text model on the page that also has a free tier. Gemini 3.6 Flash, the previous generation, carries the same per-token price, so there is no saving in staying on it.

For high-volume, simple work such as classification, translation or extraction, the Flash-Lite models are the cost floor. Gemini 2.5 Flash-Lite is the cheapest text model on the page on both input and output, and Gemini 3.1 Flash-Lite costs a little more for a newer generation. Gemini 3.5 Flash-Lite costs more than either on output, the same as Gemini 2.5 Flash, so check whether its gains justify that before you switch a pipeline to it.

Gemini 3.1 Pro Preview and Gemini 2.5 Pro double their input price above 200,000 tokens of prompt: $4 and $2.50 respectively. If your agent stuffs a whole repository into context, that threshold decides the bill more than the headline rate.

How do Batch, Flex and Priority change the price?

Google sells each main model on several service tiers, and the choice moves the price more than the choice between neighbouring models.

Batch is the easy saving: half price for any job that does not need an answer in seconds. Evaluations, nightly code review over a backlog of pull requests and bulk document processing all fit. Priority costs 1.8 times Standard on both models, so reserve it for user-facing paths where latency costs you customers.

There is also a Contributor tier on several models, at the Batch rate, where Google marks your content as used to improve its products. It is the paid-tier equivalent of the free-tier data trade: a discount in exchange for your prompts.

Context caching cuts input cost on repeated prompts. On Gemini 3.8 Flash a cached input token costs $0.075 per million, a tenth of the Standard input rate, plus $0.50 per million tokens per hour of storage. Gemini 4 Argon supports only implicit caching, at a twentieth of its Standard input rate.

What do grounding and tools add?

Grounding with Google Search on Gemini 3 and newer models includes 5,000 free search requests a month, shared across those models, then costs $14 per 1,000 requests. Google notes that one request to Gemini can trigger several searches, each billed separately, so a chatty agent can spend its monthly allowance faster than its request count suggests. Grounding with Google Maps has the same allowance and rate.

Most other tools bill as tokens. Code execution and URL context are charged at the model's token rates, File Search charges $0.15 per million tokens for embeddings plus the retrieved tokens, and agent features such as the Deep Research agent bill every intermediate reasoning token at list rates.

Want price changes like these flagged before they hit your invoice? Devshot covers them in its daily brief for developers.

What one workload costs on Gemini and Claude

To make the rates concrete, take a coding agent that sends 10 million input tokens and receives 2 million output tokens a month, all on Standard pricing. These are our arithmetic on the list prices above and on Anthropic's Claude API pricing, read on 19 October 2026.

Gemini 4 Argon's introductory price matches Claude Sonnet 5.5, and its December price matches Claude Opus 5.5. That makes the promotion a window to test Argon against Sonnet at equal cost; from December the fair comparison is with Opus. Our Claude API pricing breakdown covers Anthropic's caching and Batch multipliers in the same detail.

At the cheap end, Gemini 2.5 Flash-Lite and Claude Haiku 5.5 land within cents of each other for this workload. The difference is the free tier: Gemini's Flash-Lite models have one, and you can prototype on them before you add a card.

Costs the pricing page does not headline

• Two price rises on a timer. Gemini 4 Argon doubles on 1 December 2026 and the Gemini 3.8 and 3.6 Flash family doubles on 1 January 2027, including Batch, Flex and caching rates.
• Thinking tokens bill as output. Reasoning models can spend more on hidden thinking than on the visible answer.
• Long prompts cost more on Pro models. Above 200,000 prompt tokens, Gemini 3.1 Pro Preview and 2.5 Pro raise both input and output rates.
• Search grounding is per query, not per request. One request can trigger several billed searches after the free 5,000.
• Different prices elsewhere. Google notes that prices on its Gemini Enterprise Agent Platform may differ from this page.

Is the Gemini API worth it?

For most developers, Gemini 3.8 Flash is the best value on the page, with a free tier for prototyping, a paid rate well below Gemini 4 Argon and Gemini 3.1 Pro Preview, and Batch at half price for background jobs. Budget for its January price now, so the doubling does not land as a surprise in your cloud bill.

Gemini 4 Argon is worth testing during its promotion if your agent work needs a frontier model, because it costs no more than Claude Sonnet 5.5 until December. After that, it competes with Claude Opus 5.5 on price, and the choice should come down to your own evaluation results.

For a wider field, see our best LLM APIs ranking and the free LLM APIs guide, which covers other free tiers. If you want one key for several providers, the OpenRouter alternatives list compares routing services. Developers using Gemini through coding tools rather than the API can start with the best AI coding agents, the best free AI coding tools and the AI coding assistant pricing comparison, and Gemini CLI and the Anthropic API have their own profiles on Toolradar. If you are building on Google's stack, our Firebase pricing page covers the backend side, and the best API gateways list covers metering your own API on top.

Frequently asked questions

How much does the Gemini API cost per million tokens?

On Standard pricing, Gemini 4 Argon costs $2 input and $10 output per million tokens through 30 November 2026, then $4 and $20. Gemini 3.8 Flash costs $0.75 and $3.75 through 31 December 2026, then $1.50 and $7.50. Gemini 2.5 Flash-Lite costs $0.10 and $0.40.

Is the Gemini API free?

Partly. The free tier offers free input and output tokens on the Flash and Flash-Lite models and on Gemini 2.5 Pro, with rate limits, and Google uses free-tier content to improve its products. Gemini 4 Argon and Gemini 3.1 Pro Preview have no free tier.

Does Gemini 4 Argon have a free tier?

No. Gemini 4 Argon is paid only, at its introductory Standard rate through 30 November 2026 and twice that rate from 1 December. Batch and Flex requests cost half the Standard rate.

How much does the Gemini Batch API save?

Batch costs 50% of the Standard rate on the main models. Gemini 3.8 Flash on Batch costs $0.375 input and $1.875 output per million tokens until the end of 2026. Batch is a paid-tier feature.

Is the Gemini API cheaper than the Claude API?

At the flagship level they are priced the same until December: Gemini 4 Argon's introductory rate equals Claude Sonnet 5.5, and its December rate equals Claude Opus 5.5. At the low end, Gemini 2.5 Flash-Lite and Claude Haiku 5.5 cost about the same, but only Gemini offers a free tier.

Does Google use my Gemini API data for training?

On the free tier, Google's pricing page marks content as used to improve its products. On the paid tier it marks content as not used, except on the discounted Contributor tier, where content is used.

How we checked these prices

Every Gemini price on this page was read on 19 October 2026 on Google's Gemini API pricing page, which Google last updated on 16 October 2026, and the Claude rates on Anthropic's pricing documentation the same day. The workload costs are our arithmetic on Standard list prices. The Toolradar figures come from Google Gemini's Toolradar record, October 2026. This page reads published prices and does not benchmark model quality. Two promotional rates end on 1 December 2026 and 1 January 2027, so confirm prices on Google's page before you commit a budget.

Cite this: Devshot, "Gemini API Pricing 2026: Free Tier and Per-Token Rates", November 2026.

Devshot is the daily brief for developers who watch what their stack costs. Get it at devshot.co.

Some offers on this page may be paid placements or contain affiliate links.