Gemini

There are two things to consider: where you are located and whether you need an assistant inside Google services.

Who it is for. For those with Gmail email, Drive files, and an Android phone: Gemini is already there and reads your emails and documents on request. And for those who want a strong model for free: Google's Pro model is available without a subscription; you pay for volume (Google Help, Sep 5, 2026).

Who else. For those who work with large files: for $19.99 per month, the context window grows to a million tokens, several books in one conversation. It is also enough for images and short videos: 150 million images are created in Gemini every day (blog.google, August 2026).

Who it is not for. Personal access to Gemini CLI through Free, AI Pro and Ultra ended on Jun 18, 2026; enterprise licenses and the paid API keep working, and Google directs individual users to Antigravity.

Choosing between it and ChatGPT: see section 7.

What vibecoding isStarting from scratch: how to set tasks for agents with no coding experience

Three layers: a chat with three models, tools inside Google services, and agents that work without you.

Chat and models. The selector has three models: the fast Flash-lite, the main Flash, and the thoughtful Pro, and all three are available without a subscription. You choose response depth separately: regular, long, and Deep Think, which thinks for several minutes and is only available in Ultra. The models have version numbers in the API, and the smartest one now is Gemini 3.8 Flash (Sep 2, 2026), while Pro has been on version 3.1 since February.

Google tools. Connected apps give Gemini access to Gmail, Drive, Maps, and YouTube: "find the contract sent in March" works without uploading files. Canvas opens a document or code window next to the chat, Deep Research independently assembles a report from dozens of sources, and Notebook keeps a project with sources, up to 600 in one.

Images, video, voice. Nano Banana 2 creates images, while Nano Banana Pro edits them. Gemini Omni Flash makes videos in the app: 3-10 second clips since May 19, 2026, while the previous Veo 3.1 remains in the Flow studio and in the API. Live voice mode is only available in the mobile app, not on the web.

Agents. Gemini Spark, a personal agent that handles tasks on its own, becomes available with an AI Pro subscription; the feature list includes the Daily Brief morning summary. The feature list also includes scheduled actions. Gemini Agent and calls to local businesses are currently available only in the US: the agent in Ultra, automated website browsing in Chrome with AI Pro.

What was actually done. We have not run Gemini on a task of our own, the numbers come from Google's help pages and other people's tests. Practitioners praise the response speed and the million-token context, which iphones.ru converted into 750 thousand words, more than twice what ChatGPT holds (Jan 26, 2026).

Step 1. Open gemini.google.com and ask your first question: no registration beyond your Google account.

Computer. The macOS app launched on Apr 15, 2026: it requires Apple Silicon and macOS 15 and opens with Option + Space. There is no Windows app; use the web or Gemini in Chrome.

Three signals diverge: record audience, solid models, but not the top ones.

Audience. One billion monthly users as of Aug 11, 2026, Google's fourteenth product to reach that number and the fastest-growing in its history (blog.google, TechCrunch). It had 900 million at Google I/O in May.

LMArena. Google's best model in the text arena, Gemini 3.8 Flash, ranks eighth with 1494 points and a "preliminary" label (snapshot Sep 3, 2026); the first seven spots are held by Anthropic and Meta. In the web development arena, the best Gemini is in 16th place. Google's smartest model is now called Flash, not Pro: Pro 3.1 is fifteenth in the text arena.

Artificial Analysis index. Gemini 3.8 Flash scores 59 points versus 66 for the leader, Claude Fable 5.1 (Sep 5, 2026). Gemini 3.1 Pro scores 48 there, below Google's own Flash models.

Gemini's place among the assistants in our catalog is shown under the card title, and the full tool ranking lives on a shared page.

Models are free; you pay for limits, context, and cloud storage.

Free entry. All three models, including Pro, are available without a subscription, with 32 thousand tokens of context and a standard limit that resets every 5 hours. Deep Research and media generation use up the limit faster.

Paid tiers. AI Plus for $4.99 doubles the limit, AI Pro for $19.99 gives a limit four times higher, one million tokens of context, and the Spark agent, while AI Ultra for $99.99 and $199.99 adds Deep Think. Prices are per month; Google has no annual prices on the page (Sep 5, 2026).

For developers. The API charges $0.75 per million input tokens and $3.75 per million output tokens for Gemini 3.8 Flash, but that is the price until Dec 31, 2026; it doubles in January. Pro 3.1 costs $2 and $12, and has no free API tier.

Three questions: what Google does with chats, what the agent can do on its own, and how much to trust its answers.

Chats are used for training. History is on by default, and Google explicitly says it uses it "to improve services, including training generative models." Some chats are read by human reviewers, and those copies, disconnected from the account, are stored for up to three years and are not deleted with the history. Turn off history and chats still remain for 72 hours (Privacy Hub, updated Aug 10, 2026).

The agent makes mistakes for you. Google itself warns that Gemini "may make mistakes, for example, make a purchase or unexpectedly share your data with a third party," and asks you to monitor its actions in the browser. Google does not review third-party connected apps.

Answers need fact-checking. Practitioners complained in August 2026 about chats freezing and going blank, and refusals to generate images under safety rules. Google itself places the line "Gemini is AI and can make mistakes" below answers, so verify the numbers.

What else to know. In July 2026, publishers filed a US lawsuit alleging that Gemini was trained on their books; there has been no ruling on the merits. The service's legal entity is Google LLC, and Google Ireland in the EEA and Switzerland.

The choice is not "who is smarter," but "which ecosystem you live in."

Different environment. ChatGPT wins on tool coverage, memory across chats, and a desktop app with the Codex agent. Gemini wins because it is already built into Gmail, Drive, and Android, with one million tokens of context and a free Pro model.

Different entry price. ChatGPT has a free tier and a Plus subscription for $20, while Gemini's first step costs $4.99.

In three years, Bard became an assistant for one billion people, while the smartest model moved from the Pro line to the Flash line.