Gemini is a family of multimodal large language models (LLMs) developed by Google DeepMind, a subsidiary of Google. It serves as the successor to earlier Google models LaMDA and PaLM 2. The Gemini family includes several variants, such as Gemini Pro, Gemini Deep Think, Gemini Flash, and Gemini Flash Lite, and it powers the Gemini chatbot. The name Gemini refers to the Gemini zodiac sign and also nods to the merger of Google Brain and DeepMind, as well as NASA's Project Gemini. Announced on December 6, 2023, Gemini was designed to process multiple data types simultaneously, including text, images, audio, video, and computer code, distinguishing it from many text-only predecessors.
The development of Gemini marked a significant step in Google's efforts to compete in the generative AI space, particularly against OpenAI's GPT-4 and other large language models. Google positioned Gemini as its "largest and most capable AI model," with capabilities that extend beyond text generation to include image understanding and generation, and potential integration with robotics. The model was trained on Google's Tensor Processing Units (TPUs) and was initially available only in English, with a phased rollout across Google products and services.
Development
Google first announced Gemini during the Google I/O keynote on May 10, 2023, with CEO Sundar Pichai describing it as a more powerful successor to PaLM 2, which was also unveiled at that event. At the time, Gemini was still in early development, but it was already notable for its multimodal design, which set it apart from many existing LLMs that relied primarily on text corpora. The model was developed as a collaboration between DeepMind and Google Brain, two AI research groups that were merged into Google DeepMind in April 2023.
Demis Hassabis, CEO of DeepMind, highlighted Gemini's potential in an interview with Wired, suggesting it could surpass OpenAI's ChatGPT, which runs on GPT-4. Hassabis drew on DeepMind's experience with AlphaGo, the program that defeated Go champion Lee Sedol in 2016, and indicated that Gemini would combine the strengths of AlphaGo with other Google-DeepMind LLMs. In August 2023, The Information reported that Google aimed to launch Gemini by late 2023, with plans to integrate conversational text capabilities with AI-powered image generation, enabling contextual image creation and broader use cases.
Google co-founder Sergey Brin was brought out of retirement to assist with Gemini's development, working alongside hundreds of engineers from Google Brain and DeepMind. Brin was later credited as a "core contributor" to the project. Because Gemini was trained on transcripts of YouTube videos, Google engaged lawyers to filter potentially copyrighted material. The development process also involved extensive safety testing, as Google stated that Gemini would not be widely released until 2024 to ensure responsible deployment.
In response to Gemini's impending launch, OpenAI accelerated its work on integrating multimodal features into GPT-4. By September 2023, several companies had been granted early access to an early version of Gemini, which Google planned to offer through Google Cloud's Vertex AI service. The model was also positioned to compete with Microsoft's GitHub Copilot in the coding assistant space.
Launch
On December 6, 2023, Sundar Pichai and Demis Hassabis announced "Gemini 1.0" at a virtual press conference. The initial release included three models: Gemini Ultra, designed for "highly complex tasks"; Gemini Pro, for "a wide range of tasks"; and Gemini Nano, for "on-device tasks." At launch, Gemini Pro and Nano were integrated into Bard (Google's chatbot) and the Pixel 8 Pro smartphone, respectively. Gemini Ultra was slated to power "Bard Advanced" and become available to developers in early 2024. Google also planned to incorporate Gemini into Search, Ads, Chrome, Duet AI on Google Workspace, and AlphaCode 2.
Gemini was initially available only in English. Google touted it as its "largest and most capable AI model," designed to emulate human behavior. The company emphasized that Gemini would not be widely available until the following year due to the need for "extensive safety testing." Gemini was trained on and powered by Google's TPUs, and the name referenced the DeepMind-Google Brain merger as well as NASA's Project Gemini.
Gemini Ultra reportedly outperformed several competing models, including GPT-4, Anthropic's Claude 2, Inflection AI's Inflection-2, Meta's LLaMA 2, and xAI's Grok 1, on various industry benchmarks. Gemini Pro was said to outperform GPT-3.5. Notably, Gemini Ultra was the first language model to surpass human experts on the 57-subject Massive Multitask Language Understanding (MMLU) test, achieving a score of 90%. Gemini Pro was made available to Google Cloud customers on AI Studio and Vertex AI on December 13, 2023, and Gemini Nano was later made available to Android developers.
Hassabis also revealed that DeepMind was exploring how Gemini could be combined with robotics to enable physical interaction with the world. In line with a U.S. executive order signed by President Joe Biden in October 2023, Google stated it would share Gemini Ultra's testing results with the U.S. federal government. Similarly, Google engaged with the UK government to align with principles from the AI Safety Summit at Bletchley Park in November 2023.
In June 2025, Google introduced Gemini CLI, an open-source AI agent that brings Gemini's capabilities to the terminal, offering coding, automation, and problem-solving features with generous free usage limits for individual developers.
Updates
In January 2024, Google partnered with Samsung to integrate Gemini Nano and Gemini Pro into the Galaxy S24 smartphone lineup. The following month, Bard and Duet AI were unified under the Gemini brand, and "Gemini Advanced with Ultra 1.0" was released via a new "AI Premium" tier of the Google One subscription service. Gemini Pro also received a global launch.
In February 2024, Google launched Gemini 1.5, describing it as more powerful and capable than 1.0 Ultra. Changes included a new architecture, a mixture-of-experts approach, and a larger one-million-token context window. The same month, Google debuted Gemma, a smaller, free, and open-source range of Gemini models. Multiple publications described this as a response to Meta and others open-sourcing their AI models, and a reversal from Google's prior practice of keeping its AI proprietary.
Google announced an additional model, Gemini 1.5 Flash, on May 14, 2024, at the Google I/O keynote. At the same event, Google announced plans to integrate a desktop-optimized build of Gemini Nano directly into the Google Chrome browser via its "Built-in AI" architecture, exposing local processing capabilities to web developers through experimental APIs such as the Prompt API ("window.ai").
Two updated Gemini models, Gemini-1.5-Pro-002 and Gemini-1.5-Flash-002, were released on September 24, 2024.
On December 11, 2024, Google announced the new Gemini 2.0 Flash Experimental model. Features include a Multimodal Live API for real-time audio and video interactions, native image and controllable text-to-speech generation (with watermarking), and integrated Google Search. It also introduced improvements in reasoning and coding capabilities.
Architecture and Training
Gemini is built on a Transformer (architecture) architecture, which is the foundation of many modern Large language models. The model employs a mixture-of-experts approach in later versions, allowing it to scale efficiently by activating only relevant sub-networks for each task. Training involved massive datasets, including text, images, audio, video, and code, sourced from various repositories, including YouTube transcripts, which required careful filtering for copyright compliance.
The model was trained on Google's TPUs, which are custom Artificial intelligence accelerators. This infrastructure enabled Gemini to handle large-scale training runs, contributing to its performance on benchmarks like MMLU. The multimodal training approach allows Gemini to process and generate content across different modalities, making it versatile for applications ranging from text generation to image understanding.
Capabilities and Performance
Gemini's capabilities span multiple domains. It can perform complex reasoning tasks, generate code, understand and generate images, and process audio and video. The model's performance on industry benchmarks has been a key selling point. Gemini Ultra's 90% score on MMLU, where it outperformed human experts, was a notable achievement, highlighting its broad knowledge across 57 subjects.
In addition to benchmarks, Gemini has been integrated into various Google products, including Search, Ads, and Chrome, enhancing their AI-driven features. The model's ability to handle long context windows, up to one million tokens in Gemini 1.5, allows it to process large documents or entire codebases, making it useful for developers and researchers.
Integration and Ecosystem
Gemini is integrated into Google's cloud services, including Google Cloud and Vertex AI, allowing enterprises to access its capabilities. It also powers the Gemini chatbot, which was rebranded from Bard in February 2024. The model's availability on Android devices, via Gemini Nano, enables on-device AI processing, which is important for privacy and latency-sensitive applications.
Google has also made Gemini available through partnerships, such as with Samsung Electronics for the Galaxy S24, and through open-source releases like Gemma, which are smaller, more accessible models. The Gemini CLI, introduced in 2025, provides a command-line interface for developers, further expanding its ecosystem.
Reception and Impact
Gemini has been well-received in the AI community for its multimodal capabilities and strong benchmark performance. However, it has also faced scrutiny regarding safety, copyright issues, and the environmental impact of training large models. Google's decision to share testing results with governments reflects broader concerns about AI regulation and safety.
The launch of Gemini intensified competition in the AI industry, prompting rivals like OpenAI and Anthropic to accelerate their own developments. Gemini's integration into Google's products has also raised questions about market dominance and the ethical implications of AI deployment.
Future Directions
Google continues to iterate on Gemini, with updates like Gemini 2.0 Flash Experimental indicating a focus on real-time interactions and multimodal generation. The exploration of robotics integration suggests potential applications in physical AI systems. As of 2025, Gemini is expected to evolve further, with ongoing research into improving reasoning, efficiency, and safety.
The development of Gemini also aligns with broader trends in Generative AI, where models are becoming more multimodal and capable of handling complex tasks. Google's investment in TPUs and its partnership with Broadcom for custom chips underscore its commitment to advancing AI infrastructure.