Google's AI Gemini can be used for free. Costs come from two separate things: the paid app plans, which raise your usage limits and unlock higher models, and the API, which you call from your own programs. Their pricing works in completely different ways. This article goes through pricing, what you can do for free, the desktop apps, coding, strengths and weaknesses, and benchmarks one at a time, based on Google's official pages as of October 20, 2026. For an overview of Gemini itself, see What is Google Gemini?

Free

Chat, images, and Deep Research

The free version uses the Flash-Lite model and can generate and edit images, run Deep Research, and use Canvas. Video generation requires a paid plan.

App paid plans

From 725 yen a month

Plus 725 yen, Pro 2,900 yen, Ultra 14,500 yen and up (Japan pricing). The top model, Gemini 4 Argon, requires Pro or higher.

API

Pay per use, billed separately

Separate from the app subscription. Lighter models have a free tier, but content sent on the free tier is used to improve Google's products.

Table of Contents

1. Three ways to use Gemini — app, API, and developer tools

"The price of Gemini" depends on where you use it. Let's start by dividing it into three types.

Way to useWhat it isHow you pay
Gemini appUsed in the browser (gemini.google.com), on your phone, or in the desktop app for chat, research, image generation, and moreFree to use (monthly plans for more usage: Google AI Plus, Pro, Ultra)
Gemini APICall Gemini from your own programs or servicesPay for the tokens you use (lighter models have a free tier)
Developer tools (Antigravity)Google's coding app, CLI, and editor extensions, where AI writes code for youFree to use (Google AI Pro and Ultra raise the limits)

Joining a paid app plan does not make the API cheaper. The API is billed separately through Google AI Studio or Google Cloud.

2. App pricing plans

PlanMonthly price (Japan)Usage limitModels available (chat)Length it can read at once
Free0 yenBaselineFlash-Lite (Flash and Pro may also be used if auto selection is on)32,000 tokens
Google AI Plus725 yen2x the free planFlash-Lite, Flash, Pro128,000 tokens
Google AI Pro2,900 yen4x the free planFlash-Lite, Flash, Pro, Argon1 million tokens
Google AI Ultra14,500 yen / 32,000 yen5x / 20x the Pro limitFlash-Lite, Flash, Pro, Argon1 million tokens

Source: Google AI plans (Japan pricing), Gemini app help: Usage limits and Model availability. Confirmed October 20, 2026. The plans page lists only Flash for Plus models, but the help table shows that Pro can also be selected.

Monthly prices vary by country and region. If you use Gemini from outside Japan, check the price for your country on the plans page.

Limits are not "X times per day." According to the help page, they are calculated from the complexity of your question, the models and features you use, and the length of the conversation. They reset every 5 hours, and you can keep using the app until you reach the weekly limit. Short conversations barely reduce them, while Deep Research and video generation reduce them a lot. Even if you reach the limit on a paid plan, you can keep chatting with Flash-Lite.

The top model, Gemini 4 Argon, was added to the app on October 13, 2026, and is available only to Pro and Ultra subscribers. Plans also include cloud storage (starting at 15GB on the free plan, 400GB on Plus, 5TB on Pro, and 20TB on Ultra).

3. What you can and can't do for free

Chat and researchUses the Flash-Lite model. If auto selection is on, Flash or Pro may be used for questions that need more thought.
Image generation and editingAvailable for free (see the free column on the plans page).
Deep ResearchSearches in multiple steps and summarizes the results into a report. Available for free, but it uses up your usage quickly.
Canvas, Gems, Gemini LiveCanvas opens documents or code alongside the chat for editing, Gems save instructions, and Live lets you talk by voice. All three are listed in the free column.
Video generationFrom paid plans (Plus and up).
Reading long documentsThe free plan reads up to 32,000 tokens at once. To read long PDFs or large files in one go, you need Pro or higher (1 million tokens).
Top modelGemini 4 Argon is only available with Pro and Ultra.

Source: Google AI plans, Gemini app usage limits (confirmed October 20, 2026). For a comparison with the free versions of ChatGPT and Claude, see How far can you get with the free tiers of three services?

4. Desktop apps (Windows and Mac)

You can use Gemini from a desktop app without opening gemini.google.com in a browser. Both are free to download, and the app itself costs nothing.

MacmacOS 15 or later. Open it with Option+Space and share your screen to ask questions.
WindowsReleased on September 10, 2026. Open it from the desktop with a keyboard shortcut.

Source: The Gemini app is now on Mac and The Gemini app is now available for Windows (Google's official blog, confirmed October 20, 2026).

The features and models you can use are determined by your plan, just as in the browser version. On phones, there are apps for Android and iPhone.

5. API pricing and free tier

The API is billed by the number of tokens you use. Below are the prices per million tokens for major models (US dollars).

ModelFree tierInputOutput
Gemini 4 ArgonNone$2.00 ($4.00 from December 1)$10.00 ($20.00 from December 1)
Gemini 3.1 Pro PreviewNone$2.00 ($4.00 for prompts over 200,000 tokens)$12.00 ($18.00 for prompts over 200,000 tokens)
Gemini 3.8 FlashYes$0.75 ($1.50 from January 1, 2027)$3.75 ($7.50 from January 1, 2027)
Gemini 3.5 Flash-LiteYes$0.30$2.50
Gemini 3.1 Flash-LiteYes$0.25$1.50

Source: Gemini Developer API pricing (confirmed October 20, 2026). The lower price for Gemini 4 Argon applies until November 30, 2026, and the price for 3.8 Flash until December 31, 2026. Output prices include thinking tokens.

Free-tier content is used to improve productsOn the pricing page, the "Used to improve our products" row shows Yes for the free tier and No for the paid tier. Don't send anything on the free tier that you wouldn't want others to see.
Batch is half priceBatch and Flex, which send non-urgent work in bulk, cost half the regular price.
Answers using Google SearchFree up to 5,000 queries per month with Gemini 3 and later models. Beyond that, $14 per 1,000 queries.
Free-tier limitsCheck the per-minute and per-day request limits in Google AI Studio (the public page does not list the numbers).

Source: Gemini Developer API pricing and Rate limits (confirmed October 20, 2026).

6. Using it for coding — Antigravity

Gemini can be used for coding as well as conversation. You can write small code in the app's Canvas, but for serious work, use Google's developer tool Antigravity. It comes as an app, a CLI (used from the terminal), and editor extensions, and the AI reads and rewrites files and runs commands to get the work done.

Gemini CLI is no longer available for individuals. Google stopped providing the Gemini CLI and Gemini Code Assist extensions to free users and to Google AI Pro and Ultra users on June 18, 2026, and moved them to the successor, Antigravity CLI. The older claim that "Gemini CLI is free to use" no longer applies to individual use.

Source: An important update: Transitioning Gemini CLI to Antigravity CLI (Google for Developers Blog, May 19, 2026).

PlanAntigravity limitModels available
Free, PlusA limit that resets every weekGemini 3.8 Flash and others
Google AI ProResets every 5 hours (up to the weekly limit), with a higher weekly limit tooGemini 4 Argon, Claude Opus 5.5, Sonnet 5.5, and more
Google AI UltraThe largestSame as Pro (plus models from other companies)

Source: Antigravity documentation: Plans and FAQ (confirmed October 20, 2026). The specific number of requests is not published. It is available in Japan, but users must be 18 or older.

Code completion (Tab) has no limit on any plan. On Pro and Ultra, you can choose Claude models as well as Gemini inside Antigravity.

7. Strengths and weaknesses, and benchmarks

Numbers published by Google

Google has published a table comparing its top model, Gemini 4 Argon, with GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 (Google's own numbers). Some of the results are excerpted below.

EvaluationGemini 4 ArgonGPT-6 AstraClaude Fable 5.1Claude Opus 5.5
DeepSWE v1.1 (coding)77.9%74.1%67.4%74.2%
Terminal-bench 4.0 (work in the terminal)57.4%58.2%57.9%66.4%
LVBench (understanding long videos)91.7%87.5%79.7%83.7%
Vals Index (work tasks)68.9%63.1%65.8%67.0%

Source: Gemini (Google DeepMind, confirmed October 20, 2026). The thinking settings are not listed in the table.

Third-party evaluations

Source: Artificial Analysis models list and Gemini 4 Argon page (confirmed October 20, 2026). Values may change over time.

Strengths and weaknesses

Strength: smart for the priceIn third-party overall scores, Argon ties GPT-6 Astra at 53 points, and the cost of one run is about 60% of Astra's and about one-third of Claude Opus 5.5's.
Strength: long videos and long documentsGoogle's announcement says it outperformed the three compared models on long video understanding (LVBench). In the app, it is available with Pro and above, and in the API it can read up to 1 million tokens at once.
Strength: integration with Google servicesConnects with Gmail, Google Drive, and Google Search (see What is Google Gemini? for details).
Weakness: long work in the terminalEven in Google's own table, Argon falls about 9 points short of Claude Opus 5.5 on Terminal-bench 4.0.
Weakness: response speedIn Artificial Analysis's measurements, Argon outputs about 84 tokens per second. Among the models compared, it is not one of the faster ones.
The free version is differentThe numbers above are for Argon and Flash. The free app usually answers with Flash-Lite, so you won't get the same performance.

8. Which one should you use?

  1. Try the free app first For chat, research, and image generation, the free version is often enough. On a computer, the Windows and Mac apps are convenient.
  2. If you use it every day for long documents or work, choose Pro At 2,900 yen a month, you can read up to 1 million tokens and also use the top model, Argon. Plus (725 yen) is for people who only want double the usage.
  3. Use Antigravity for coding You can try it for free. If you use it daily, choose Pro to raise the limits.
  4. Use the API if you're building it into your own service Lighter models can be tried on the free tier, but content sent on the free tier is used to improve products. Use the paid tier for production.

For how to split the work between ChatGPT, Claude, and Gemini, see How to choose between ChatGPT, Claude, and Gemini. If the app shows "Something went wrong (13)," see How to fix Error 13.

Summary

  • Gemini can be used for free. Paid options are the app plans (from 725 yen, 2,900 yen, and 14,500 yen a month) and the API, which you pay for by usage. The two work separately.
  • The free version uses the Flash-Lite model and includes image generation, Deep Research, and Canvas. Video generation and the top model, Argon, require a paid plan.
  • Limits are based on computation rather than a number of uses. They reset every 5 hours, and you can use the app until you reach the weekly limit.
  • Desktop apps are available for both Windows and Mac, and can be downloaded for free.
  • The API has a free tier for lighter models, but content sent on the free tier is used to improve products.
  • For coding, use Antigravity. The individual Gemini CLI ended on June 18, 2026.
  • In third-party overall scores, Argon gets 53 points. It is smart for the price and strong with long videos, but for long terminal work it falls short of Claude Opus 5.5.

FAQ

Q. Can I use Gemini for free forever?

A. As of October 20, 2026, you can use it for free with a Google account. The free version's limits reset every 5 hours, and you can use it until you reach the weekly limit.

Q. If I subscribe to a paid plan, can I use the API without limits?

A. No. App plan pricing and API pricing are separate. The API is billed by usage through Google AI Studio or Google Cloud.

Q. Does answer quality differ between the free and paid versions?

A. Yes. The free version usually answers with Flash-Lite. Plus lets you choose Flash and Pro, and Pro and Ultra also let you choose the top model, Argon. The length it can read at once also differs, from 32,000 tokens on the free plan to 1 million tokens on Pro and above.

Q. Is it safe to enter work information?

A. It depends on how you use it. The pricing page states that content on the API free tier is used to improve products. If you use Gemini at work, check your company's rules and how your work Google account handles data before you start.

Sources