Gemini, Google’s large language model (LLM), has come a long way since it was introduced as Bard.Google’s experimental ChatGPT competitorhas become central to Google’s identity. What was once Duet AI is now Gemini for Google Workspace. Gemini has also taken over Google Assistant on thebest Pixel phones, showing Google’s effort to consolidate its artificial intelligence technologies under the Gemini umbrella. Gemini pulls together several products, and we explain what it is, how it works, and what to expect.
What are large language models?
Large language models (LLMs) are the basis for AI chatbots and much more. Here’s what’s going on behind the scenes
What is the Google Gemini chatbot?
Gemini is Google’s evolution of Bard
On June 20, 2025, Google announced a major rebranding of Bard, its experimental AI chatbot. The tool isnow known as Geminiand has kept its core features. Geminiintroduced advanced skillsin reasoning, planning, and understanding, allowing it to handle complex summarizing and coding tasks while providing better context-aware responses.
Gemini is your interface for accessing Google’s LLM and generative AI, like ChatGPT. Initially, Gemini delivered natural, text-based responses. However, like the competition, itadded generative AI featuresto its toolkit.
What is generative AI?
An agent of the human will, an amplifier of human cognition. Discover the power of generative AI
Gemini is a free product.Gemini Advancedis available through a subscription. It includes more features and gives more accurate answers. It’s part of the newly added tier to the Google One plan for $26 per month or $20 if you pay yearly. On top of access to a better AI model, it features 2TB of Google Drive storage, which costs $10 per month, plusmore Google One features. Gemini Advanced is available in more than45 languagesand over150 countries and territories. It will roll out to more regions and languages in the future.
Gemini can also be fullyintegrated with Google Workspace, offering AI-driven support for writing summaries, data analysis, and image generation, much like Duet AI did in Gmail, Docs, and Sheets. Access to these featuresrequires a subscription, which is an additional $20 per user per month.
Google also released a Gemini app for Android, which was never available for the older chatbot version. After installing Gemini on your phone,Gemini replaces Google Assistantas the assistant on your device. This unlocksnew featureson your phone. The familiar “Hey Google” opens up interaction with Gemini, and its screen awareness allows it to generate text or answers based on the visible content.
What are AI repetition penalties?
Here’s your remedy when AI gives you the wrong answers
Gemini is a family of versatile Google AI models
The AI model behind Gemini (the chatbot) is also named Gemini. The Gemini family is a collection of capable AI models, with each version crafted for different purposes and performance levels. The lineup includes Gemini Nano, Gemini Pro, Gemini Flash, and Gemini Ultra. These models run on most devices, which is why Google is deploying them everywhere.
Gemini Nano
Gemini Nano is the mobile-optimized member of Google’s Gemini AI lineup.First introducedwith theGoogle Pixel 8 Proin the December Feature Drop,Gemini Nanosupports on-device processing to handle privacy-sensitive functions like suggesting replies in encrypted messaging apps without sending data elsewhere. Processing on mobile devices also means low latency, real-time performance, and accessibility to many features,even offline.
Gemini Ultra
Gemini Ultra is the most powerful model in Google’s Gemini AI lineup, designed to handle the most complex and demanding tasks. As the flagship model,Gemini Ultraexcels in multimodal reasoning, allowing it to simultaneously process and understand multiple forms of input, such as text, images, audio, or code. The model also shines in areas like mathematics and physics.
Gemini Pro
Gemini Pro is Google’s go-to AI model, built to perform well across various tasks. This is the brain behind the Gemini chatbot and Workspace apps. The recent Gemini 1.5 Pro works with a context window of up to two million tokens, the longest for any large-scale model,according to Google. This allowsGemini 1.5 Proto handle massive documents, thousands of lines of code, and hours of media for tackling complex challenges.
Gemini Flash
Gemini 1.5 Flash is the latest addition to Google’s Gemini AI family. It’s a lighter, faster, and more budget-friendly option than the more powerful Gemini 1.5 Pro. Despite its streamlined design,its one-million-token contextwindow allows it to handle complex tasks.
Gemini has different versions, like Ultra, Nano, Flash, and Pro. Each version gives you more features or improvements. Even though Gemini Pro 1.5 is smaller than Ultra 1.0, it’s better in some ways because it’s more up-to-date.It beats the Ultra 1.0 modelon 16 of 19 text benchmarks and 18 of 21 vision benchmarks.
Is Google Gemini a chatbot? Can it create content?
Gemini can create content, but it is more ambitious than a chatbot. Gemini is amachine learningframework. It’s taught by entering human stuff (online content, generally) into it and helping it make rules to understand that content. Do that enough, and LLMs can process language data to put together sentences and mimic certain styles likeChatGPTand Bard. They’re like expert puzzle solvers creating mathematical ways to “solve” human speech. The more they learn, the better they get.
Most LLMs specialize in only a few things, like speech or images. That keeps them focused and reduces the enormous resources they require. Google is skilled at creating efficient AI models that are deeply trained on a limited array of content, contrasting OpenAI’s system of throwing almost everything it can at the AI.
What is OpenAI?
OpenAI is igniting the AI revolution with bold projects and visionary alliances
Gemini appears to differ from the existing models because it’s been trained asmultimodal from the beginning. Multimodal means the AI learns and creates all kinds of content, not just one “language.” Gemini handles speech, matches, reasoning problems, code, images (including emojis), video, audio, and more. It’s like the polymath or Renaissance Man of the LLM world.
As you’re able to see from the examples, that seems to make Gemini very good at understanding context and interpreting that information correctly for users, regardless of the subject.
Gemini appears to be very good at what it does within its scope.It scored 90%on the Massive Multitask Language Understanding (MMLU) test, which is better than most human language experts and in line with Google’s past performance.
Google also saysGemini beats existing AI models in 30 of 32 academic tests used to score LLMs. However,other reports saythat Gemini Pro can beat GPT-3.5 (which powered much of the ChatGPT content we’ve seen this year) but is beaten by the newer GPT-4, while Gemini Ultra narrowly beats GPT-4.
No AI is as multimodal as Gemini. Businesses that use this trained AI can adapt it to nearly anything. That holds value for companies wanting to customize AI services to do anything from recognizing counterfeit handbags to imitating a helpful Swedish uncle on a customer service chat. Google also mentions a few other possibilities, such as:
What can Gemini do for the everyday consumer?
Gemini supports a range of AI features across multiple channels. Like ChatGPT, it has strong generative capabilities. Tell Gemini you’re planning a party, and it helps you with a shopping list or theme ideas. Need a recipe? Gemini guides you through the cooking process step by step. Because it’s a multimodal LLM, it works with different input types, like text, code, audio, images, and videos.
Snap a photo of a plant, and Gemini identifies it and offers care instructions. It’s also integrated with the Google ecosystem. When planning a road trip, it puts together a playlist on YouTube or suggests the best routes on Google Maps.
Is Google Gemini different from Google Bard?
Yes. Gemini differs from Google Bard, but a little context makes this answer less confusing. Until February 2024, Google Bard was the user interface Google used with its LLMs. The original Bard,launched in early 2023, was an earlier attempt at consumer-facing AI (in the context of these early 2020s AI LLMs, several months can be a long time).
When it launched in March 2023,Bard used Google’s LaMDA(Language Model for Dialogue Applications) model. A few months later, Bard got its first major update with therelease of PaLM 2 at Google I/O. In December 2023, Google gave Bard its biggest update with the switch to the Gemini Pro model. In February 2024, the Bard brand was discontinued, with the interface now called Gemini.
What’s the deal with PaLM 2 now that Gemini has been released?
It’s complicated, and we don’t have a good look behind the scenes.PaLM 2 was a massive updateto Google’s language-focused LLM made earlier in 2023. PaLM 2 excels at language tasks like translation. While Google made PaLM 2 modules that handle other things like reading medical scans, it isn’t multimodal like Gemini. However, it provides lightweight AI services for businesses that want to build their own AIs by tapping into the work Google has done, using theGoogle Vertex AI platform, which Gemini 1.5 Pro is also on.
Gemini and PaLM 2 don’t appear to be competitors, and Gemini is the model most people will interact with when using AI products and hardware.Google DeepMind, formed by merging the two previous projects,Brain Teamand DeepMind, is in charge of both. Google refers to PaLM 2 and Gemini as separate AI models with a different focus, though they may work together for certain tasks.
How to use Gemini in your workflow
If you want to use the user-facing version of Google Gemini,visit the Gemini websiteor download the Gemini app on your Android phone. On the Apple iPhone, Gemini is available within the regular Google app.
If you’re a developer interested in using the underlying AI model for your projects, stop byDeepMind’s web page for Geminiand look for a sign-up option to learn more or a sign-in option for your developer account to get started with the Gemini API kit. From there, you can incorporate Gemini services into your apps based on Gemini models that fit your needs.
Let Gemini plan your summer vacation: Here’s how the generative AI platform can help
Use Gemini for a stress-free trip-planning experience
How much does Gemini cost?
For developers and companies using the underlying Gemini AI model, specific Gemini pricing is difficult to parse. We suggest looking at Google Vertex andits pricingfor gen AI services, which vary based on the type of content and the service a business is interested in.
Is Google Gemini safe?
DeepMind says that Gemini was trained with safety in mind and will bedeployed responsibly. Google is vague about what that entails, but it likely means that Gemini won’t do anything too naughty, invasive, or illegal.
Left largely untouched is the question of how Gemini is consuming our content, proprietary work, and conversations, as well as how it could take jobs, make money in unethical ways, or exploit vulnerable groups. Those are questions raised about all LLMs, and we have more questions than answers.
When you converse with Google Gemini, your words may be used to train the AI.Your conversations could be auditedand reviewed by Google workers tasked with improving the product, as prominently disclosed when you first open Gemini. Be mindful of what you share with the AI, and don’t give out private information you wouldn’t be comfortable saying out loud elsewhere on the internet.
The AI Race
It looks like Gemini has started to close the gap with its AI rival, GPT. While we can only guess what OpenAI’s next version of GPT will bring, the competition between these two giants is heating up. Recent developments highlight how intense this AI race has become.Samsung and Google joined forcesto bring AI tools toflagship Android phones. Meanwhile,OpenAI teamed up with Apple, integrating its AI into the newApple Intelligence platformon iOS. As each company continues to push the envelope, the stakes get higher. The future of AI is up in the air, but one thing’s for sure: this race is far from over.
