News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
imgwavvidGemini

Gemini

Google DeepMind's multimodal LLM family, from Nano to Ultra

Gemini is a family of multimodal large language models from Google DeepMind, announced on 6 December 2023 as the successor to LaMDA and PaLM 2. It ranges from on-device Nano to the complex-task Ultra, powers the Gemini chatbot, and has grown through versions 1.5, 2.0, 2.5 and 3.

Share on WhatsApp

  • Announced 6 December 2023 as successor to LaMDA and PaLM 2
  • First model reported to beat human experts on MMLU, scoring 90%
  • Gemini 1.5 added a mixture-of-experts design and a one-million-token window
  • Nano Banana drew 10 million new users and 200 million image edits in weeks
  • Gemma, a small open range derived from Gemini, launched in February 2024

What it is

Gemini is a family of multimodal large language models built by Google DeepMind, announced on 6 December 2023. It replaced LaMDA and PaLM 2 and comes in tiers: Ultra for highly complex tasks, Pro for a wide range of work, and Nano for on-device use. Later lines added Deep Think, Flash and Flash Lite. The models handle text, images, audio, video and code together in a single context window, and they power the Gemini chatbot. The name nods to the Latin word for twins, marking the joining of Google Brain and DeepMind, and to NASA's Project Gemini.

How it came about

Google first described Gemini at its I/O keynote on 10 May 2023, when CEO Sundar Pichai said the project was still early. It was built jointly by DeepMind and Google Brain after those groups merged. Sergey Brin came back from retirement to help, and was later named a core contributor. A roadmap reported in August 2023 targeted a late-2023 launch, and some firms got early access. Gemini 1.0 was announced on 6 December 2023 by Pichai and Demis Hassabis, with Pro and Nano shipping first and Ultra following in early 2024.

Inside the model

Gemini is multimodal by design rather than text-only. A single context window can mix text, pictures, video and audio in any order, and the model can answer in the same mixed way. Images arrive at varying resolutions, video is treated as a run of frames, and audio is sampled at 16 kHz then turned into tokens by the Universal Speech Model. Training data covers web documents, books, code, images, audio and video, in many languages. The models are decoder-only transformers tuned for Google's Tensor Processing Units; the 1.0 line uses multi-query attention, and 1.5 introduced a mixture-of-experts design with a one-million-token window.

What it can do

At launch Google said Gemini Ultra beat GPT-4, Claude 2, Inflection-2, LLaMA 2 and Grok 1 on several industry benchmarks, while Pro beat GPT-3.5. Ultra was reported as the first model to exceed human experts on the 57-subject MMLU test, scoring 90%. Gemini 2.5 Pro Experimental reached the top of the LMArena human-preference leaderboard and posted strong results on Humanity's Last Exam, with reviewers noting better reasoning. Published scores also showed cases where models from Anthropic, xAI and OpenAI led. Scholars warned that benchmark results are hard to read without knowing the training data.

How to use it

Gemini Pro reached Google Cloud users on AI Studio and Vertex AI on 13 December 2023, and Nano went to Android developers. From February 2024 the Gemini brand covered the chatbot and a paid Google One tier, with a global Pro launch. Gemma, a small open range derived from Gemini, appeared the same month. Gemini CLI, an open-source terminal agent, arrived in June 2025. Image models Nano Banana and Nano Banana Pro are offered in the Gemini app, Google AI Studio and Vertex AI.

Safety and testing

Google delayed wide release of Gemini 1.0 to allow extensive safety testing. In line with a US executive order from October, the company said it would share Ultra test results with the federal government, and it held talks with the UK government to follow the principles agreed at the AI Safety Summit at Bletchley Park. Lawyers were used during development to filter copyrighted material from YouTube transcripts. Image outputs carry SynthID, an invisible signature marking AI-generated content.

Reception and debate

MIT Technology Review called the pre-launch speculation peak AI hype, and Business Insider framed Gemini as a make-or-break moment for Google. Oren Etzioni expected an arms race with OpenAI, while Alexei Efros welcomed the multimodal approach. Google shares rose 5.3% the day after the unveiling. The company drew criticism for a demo video presented as a live interaction that had been edited, though disclaimers outside the video said it was not recorded in real time. A Nano Banana tie-up with Google Earth was withdrawn within 24 hours in July 2026 after researchers and users showed it could produce misleading disaster and military imagery.

Why it matters

Gemini became the model family behind Google's chatbot and was folded into Search, Ads, Chrome, Workspace and Pixel devices, and into Samsung's Galaxy S24 line from January 2024. The smaller Gemma range, released in February 2024, was read as a shift toward open weights after years of proprietary models. Nano Banana drew more than 10 million new app users and over 200 million image edits within weeks, and its figurine trend spread across Instagram and X. Gemini Robotics, a vision-language-action model, followed in March 2025.

Questions people ask about Gemini

What is Gemini?

A family of multimodal large language models from Google DeepMind, announced on 6 December 2023. It handles text, images, audio, video and code in one context window, and comes in tiers from on-device Nano to Ultra for complex tasks. It powers the Gemini chatbot.

Who built it and when?

Google DeepMind, working with Google Brain after the two merged. It was previewed at Google I/O on 10 May 2023 and launched as Gemini 1.0 on 6 December 2023, with Ultra following in early 2024. Sergey Brin returned to help and was credited as a core contributor.

How is it different from earlier Google models?

It replaced LaMDA and PaLM 2 and was trained on mixed data types rather than text alone. Gemini 1.5 added a mixture-of-experts architecture and a one-million-token context window, a large jump over the 1.0 generation.

How can I use it?

Through the Gemini chatbot and app, Google Cloud services such as AI Studio and Vertex AI, and Android devices running Gemini Nano. The open-source Gemini CLI terminal agent arrived in June 2025, and the smaller Gemma models are openly released.

What is Nano Banana?

The nickname for Gemini 2.5 Flash Image, an image generation and editing model tested anonymously on LMArena from August 2025 and released on 26 August 2025. It supports subject consistency across edits, multi-image fusion and SynthID watermarking. Nano Banana Pro followed on 20 November 2025.

Good reading and sources

Where this page comes from

The facts on this page come from the Wikipedia article "Gemini (language model)", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryLLM rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.

It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

The weekly digest

Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.