What is Gemini? Google's AI Explained

Gemini in 30 seconds

Gemini is Google's AI assistant and the family of language models behind it. It is multimodal from the ground up — text, images, audio, video and code in one model — and it replaced Bard in early 2024. There is a free tier, and paid tiers start at roughly 5 EUR a month.

What sets Gemini apart from ChatGPT and Claude is not primarily the model, but where it sits. Gemini is built into Gmail, Docs, Android, Chrome and Google Search — tools many people already have open all day. If your work already lives in Google Workspace, that is the difference that decides it, not who wins on benchmarks.

What does Gemini mean?

Gemini is Latin for "the twins" — the same word as the star sign. That is why searches for "gemini" today mix two entirely different things: Google's AI and horoscopes.

This article is about Google's AI only.

From Bard to Gemini

Google launched its first public chatbot as Bard in March 2023, a fast answer to ChatGPT. The reception was lukewarm.

In February 2024 the name changed to Gemini. It was not just a rename: Bard ran on older models retrofitted for chat, while Gemini was built multimodal from the ground up. Bard no longer exists in any form — old links to bard.google.com now point to Gemini.

The Gemini models in 2026

Gemini is not one model but a family. The names change often, so here is where things stand today (checked 2026-09-19 against Google DeepMind's own model page):

• Gemini 3.1 Pro — the flagship for complex tasks, reasoning and creative work. This is what the paid tiers give you.
• Gemini 3.8 Flash — built for agentic tasks at scale, meaning multi-step jobs where the model works on its own over time.
• Gemini 3.5 Flash-Lite — the fast, cheap variant for high-volume work where cost per call matters more than absolute quality.
• Gemini 3.1 Deep Think — a dedicated reasoning mode for hard problems in science, research and engineering.

Around the core models sits a set of specialised ones: Gemini Image, better known by its codename Nano Banana, for image generation and editing; Gemini Omni for video and other multimodal creation; Gemini Audio for real-time audio; Gemini Robotics; and Gemini Embedding 2 for semantic search.

A practical note: do not worry too much about version numbers. Google changes them faster than anyone can update their articles, and the difference between two adjacent versions rarely shows in everyday use. What actually matters is which tier you pay for, because that decides which model you get and how much you may use it.

Deep Think and Deep Research — what is the difference?

The two names are close enough that they get confused constantly, but they do completely different things:

• Deep Think is a reasoning mode. Instead of answering immediately, the model explores several hypotheses in parallel and works through the problem step by step. It is built for maths, science and engineering — problems where a right answer exists but is hard to reach. Deep Think requires the most expensive tier.
• Deep Research is a research mode. The model searches the web for you, reads a large number of sources and assembles a report with references. It takes minutes, not seconds, and suits getting an overview of a topic you do not already know.

Put simply: Deep Think thinks longer, Deep Research reads more. If you need one, the other rarely helps.

What does Gemini cost?

This is the question we get most often, and the hardest one to find a straight answer to. Prices below are from Google's Swedish pricing page, checked 2026-09-19, converted here for reference — Google changes them periodically, so always check Google's own pricing page before committing.

• Free — the Gemini app, the web version and the Flash model with standard usage limits. For most personal use that goes further than people expect.
• Google AI Plus — 55 SEK/month. Doubled limits on the Flash model, video generation, Gemini Notebook (formerly NotebookLM), Gmail and Docs integration, and 400 GB of storage.
• Google AI Pro — 255 SEK/month. This is where Gemini 3.1 Pro arrives with four times the limits, plus Deep Research, Gemini Pro in Google Search, more credits in Google Flow and 5 TB of storage. Most people using Gemini for work land here.
• Google AI Ultra — from 999 SEK/month, and 2,599 SEK/month for the top tier. Up to twenty times the limits, access to Deep Think, Gemini Spark, Project Genie and YouTube Premium.

Worth noting: that covers Gemini as a finished product. If you build something against the API you pay per token instead, exactly as with OpenAI and Anthropic, and the maths looks entirely different.

Does Gemini work in languages other than English?

Yes. Gemini understands and answers in Swedish and most other major languages, and has done since broad language support rolled out in 2024. In practice the gap to English shows up in three places:

• Nuanced prose is often better in English. If you are writing something for publication it can pay to work in English and translate last.
• Local abbreviations, agency names and domain jargon are less reliable. Check them.
• Voice mode works but with noticeably less natural pronunciation than in English.

This applies to all the large models, not just Gemini. Smaller languages are a small share of the training data, and it shows.

Where you will find Gemini

Gemini's real strength is its reach. You meet it in more places than you think:

• The Gemini app and gemini.google.com — the chat interface. You sign in with your normal Google account; no separate account is needed.
• Google Workspace — in Gmail (writing and summarising email), Docs (drafts and rewrites), Sheets (formulas and analysis) and Meet (meeting notes).
• Android — Gemini has replaced Google Assistant as the built-in assistant on newer phones.
• Chrome — AI help directly in the browser, against the page you have open.
• Google Search — through AI Overviews, the summaries now sitting on top of many result pages. This is probably the Gemini feature most people use without knowing it.

For businesses that last point matters in a different way: AI Overviews change how much traffic a search actually sends on to websites. It is one of the things we work on in our AI consulting.

Nano Banana — the Gemini image model

During 2025 Google's image model picked up a codename that stuck far better than the official one: Nano Banana. It was first used while the model was being tested anonymously, and became established enough that Google now uses it too.

The model is good at something image generators long struggled with: editing an existing image without redoing all of it. You can ask it to swap a background, change a garment or remove an object, and the rest of the image stays put. It also handles text inside images far better than earlier generations.

Image generation is included in the free tier in limited quantity, and at higher volume in the paid tiers.

How to use it, how Nano Banana 2, 2.1 and Pro differ and which prompts give the best images is covered in Nano Banana 2 and 2.1 — how to use Google's AI image generator.

Gemini, ChatGPT or Claude?

The honest version: the differences between the three are smaller than the marketing suggests, and they trade places several times a year. Choose by where you already work, not by the latest benchmark.

• Gemini if you live in Google Workspace. The integration is hard to beat, and Deep Research is strong.
• ChatGPT if you want the broadest ecosystem of add-ons, tools and ready-made GPTs.
• Claude if you work with long documents, careful analysis or code.

We have tested the models against each other on the same task in our comparison of AI chatbots, and covered OpenAI's current tiers in GPT-5.6: Sol, Terra and Luna.

Getting started

1. Go to gemini.google.com or download the Gemini app and sign in with your Google account.
2. Start on the free tier. Try it for at least a week on real work before paying anything.
3. Connect what you already use — Gmail and Docs — so the model has something to work with.
4. Upgrade only when you actually hit the limits, or when you need Deep Research.

Limitations worth knowing

Gemini is a language model, with the same underlying weaknesses as every other one:

• Hallucinations. It can invent sources, figures and quotes that look entirely credible. Verify anything that matters.
• Confidentiality. Read the terms before sensitive material goes into a chat, especially on the free tier. Company data is covered by different terms through Workspace than through a personal account.
• The integration is also lock-in. What makes Gemini practical — that it sits in everything — also makes it more expensive to leave later.
• Version confusion. Google runs different model versions on different surfaces at the same time, so the answer you get in Search need not come from the same model as the app.

If multimodal, token and reasoning mode are new terms to you, we cover the basics in what an LLM is.