In short
In short: Gemini is Google's AI model - the program that understands and composes text, images and speech, just like ChatGPT (OpenAI's chatbot) or Claude (Anthropic's assistant, the one I'm writing this text with). The difference is that Google doesn't sell it as a separate app you have to download and open on purpose. It puts it straight into the things you already use every day - your search engine, your phone, your inbox. You don't ask whether you want it, it's simply already there.
Technically Gemini is a family of models, not a single product. There's a version for quick chat in the browser, a version built into Android (the operating system most phones in the world run on, aside from iPhones), and a version for businesses through Google Cloud. The big release Gemini 3 is the one Google pushed everywhere at once - from Gmail (Google's mail service) to search.
The next generation is already here. On October 13, 2026 Google released Gemini 4 Argon globally, but not free for everyone. Through the API it is on the paid tier only, with a starting price of 2 dollars per million input tokens and 10 per million output tokens. In the Gemini app and in AI Mode in search it is for Pro and Ultra subscribers. The free tier and users without a subscription are not listed in the announcement. The results Google cites for the model are its own.
Two weeks later, on October 28, 2026, came Gemini 4 Carbon - in Google's words its strongest model, with leading cybersecurity capabilities. It started with a limited group of cyber defenders, then with trusted partners through the Fairwind program. On November 10, 2026 Google released it for everyone: developers, businesses, scientists and consumers. The price didn't move - the introductory price is 2 dollars per million input tokens and 10 per million output tokens, the same as Argon. All the numbers for the model are Google's own.
On the same date, November 10, 2026, Google Cloud announced Gemini for healthcare - a solution that in the company's words brings together fragmented clinical, operational and administrative systems in hospitals. According to Google, clinicians lose close to 28 hours every week to documentation and administrative tasks. It is initially available in preview with launch partners - HCA Healthcare, Highmark Health and Northwell Health. Preview, not general availability, and the 28-hour figure is Google's, with no described method.
On November 18, 2026 Gemini 4 Neon joined the family, next to Argon, Carbon and Hydrogen. The price is 3 dollars per million input tokens and 15 per million output tokens, with a 1 million token context window. What matters for the bill is something else: according to Google, at the default thinking level, high, Neon comes out 32% cheaper per task than Carbon. In Google Antigravity it can work on its own or conduct parallel Hydrogen subagents: one conducts, the others execute. Both the cost per task and the rankings are Google's, not an independent check.
On November 19, 2026 Google released Tool Retrieval in preview in the Gemini Interactions API. The problem it solves: an agent wired to dozens or even hundreds of tools sends the schema of every one of them with every request, even when the task needs one. Now the library stays on standby. You add tool_search to the request and defer_loading: true on the tools you want to hold back; the model still sees their name and description, but the parameter schemas wait outside the context until the task asks for them. There's no index to build or host, Gemini runs the search on the server, and billing is at the standard rates for the model you call. It works with Gemini 3.6 and later, including all Gemini 4 models.
On the same day, November 19, 2026, Genie 4 became available to developers through the Gemini API and Google AI Studio, a day after Genie 4 itself launched. From a text prompt, a reference image, or both together, it generates continuous 720p worlds at 24 frames per second with synchronized audio, and with keyboard controls you can change the weather, the lighting and the objects on the fly. Every output carries an imperceptible SynthID watermark. Both announcements are Google's, not an independent check.
In practice it looks like this: you ask your inbox "who gave me a quote for the bathroom last year" and Gemini reads your emails and answers you in a plain sentence, not a link. On Android the model acts proactively - it books you a spin class ticket on its own, or finds the syllabus in an email and adds the required books to your cart. In May 2026 Google also launched "Gemini Spark" - an agent you literally email a task to, and it does it on its own through Chrome, while you watch the progress from your phone.
The reason Gemini exists is simple - Google has three billion phones, the most used search engine in the world, and all of your mail and documents. When ChatGPT showed that people want to ask AI directly instead of searching and clicking links, Google felt a threat to its own search engine. The answer wasn't a new app to download, it was the opposite - letting the AI-driven assistant disappear into the environment you already live in.
You don't download Gemini. It's already in your phone, reading your mail and deciding for you - the question is whether anyone ever asked you.
Here's Where the Real Thing Starts
There's no mystery here - the owner is Google, that is, Alphabet. But that's exactly why Gemini is a different case from ChatGPT or Claude. Gemini is built into your phone's operating system, into your mail, into your calendar. Turning it off means deciding to live without the core services you've used for years.
So for me the question isn't whether Gemini is a good model - it probably is. The question is who decides when it reads my mail and acts in my name, and whether I have any choice at all, once it's already there by default.
Primary sources
Google: Introducing Gemini (6 December 2023)Google - Gemini 4 Argon: Built to solve your hardest problems, 13.10.2026Google - Introducing Gemini 4 Carbon, 28.10.2026Google - Introducing Gemini 4 Neon, 18.11.2026Google - Introducing Tool Retrieval in the Gemini API, 19.11.2026Google - Genie 4 now available through the Gemini API, 19.11.2026The visual is generated code-art, no third-party images.