Google has officially unveiled Gemini 2.0, its most advanced large language model (LLM) designed, it claims, for the “agentic era.” The model has been trained to offer enhanced capabilities, including multimodal output with image and audio generation, as well as seamless integration with tools like Google Search and Google Maps.
Developers can now access Gemini 2.0 Flash, an experimental version, via the Gemini API on Google AI Studio and Vertex AI, while Gemini and Gemini Advanced users can explore a chat-optimised version through the desktop model dropdown.
“2.0’s advances are underpinned by decade-long investments in our differentiated full-stack approach to AI innovation,” wrote Google and Alphabet CEO Sundar Pichai in an official blog post. “It’s built on custom hardware like Trillium, our sixth-generation TPUs. TPUs powered 100% of Gemini 2.0 training and inference, and today Trillium is generally available to customers so they can build with it too. If Gemini 1.0 was about organising and understanding information, Gemini 2.0 is about making it much more useful.”
Google is simultaneously advancing its AI research through Gemini 2.0, including projects like Astra, a universal AI assistant prototype, Mariner, a Chrome extension capable of action-taking, and Jules, an AI-powered coding assistant. The company emphasised its commitment to safety, noting that trusted testers are shaping experimental features.
Gemini 2.0 Flash delivers faster performance and improved benchmark results, outperforming Gemini 1.5 Pro on key metrics, such as a 92.9% score on the Natural2Code benchmark for code generation, compared to Gemini 1.5 Pro’s 85.4%. Google claims that the model also excels in mathematical problem-solving and long-context understanding, broadening its potential applications. Its multimodal capabilities allow it to process and generate images, audio, and video, making it a versatile platform for developers.
Gemini 2.0 is already being tested within Google products, with AI Overviews in Search as an early use case. General availability is planned for January, with more model sizes and integration expected in 2024.
A key feature of Gemini 2.0 is Deep Research, an advanced tool that acts as a virtual research assistant. It compiles detailed reports on complex subjects using the model’s reasoning and long-context capabilities. This feature is currently available to Gemini Advanced users.
Additionally, Google has launched a new Multimodal Live API, enabling real-time audio and video input alongside tool integration for interactive applications. This is aimed at developers building dynamic experiences.
Gemini 2.0 also powers several research prototypes, including Project Astra for AI assistants, Project Mariner for human-agent interaction via browser extensions, and Jules for coding assistance. These projects remain in the exploratory phase, with early feedback from trusted testers helping to refine future updates.
Users can currently access the chat-optimised version of Gemini 2.0 on both desktop and mobile web platforms, with plans to expand to the mobile app soon.
Competitive landscape and market reception
Google’s Gemini 2.0 enters a highly competitive field, facing strong rivals such as OpenAI’s GPT-4, Microsoft’s Copilot, and Anthropic’s Claude. OpenAI’s GPT-4 remains a dominant player, known for its advanced text generation and reasoning capabilities. Microsoft’s Copilot, integrated across the Office suite, focuses on enhancing productivity, while Anthropic’s Claude prioritises safety and ethical considerations in AI interactions.
Early reviews of Gemini 2.0 highlight its standout multimodal features, including native image and audio generation, and its integration with Google Search and Maps. These capabilities position it as a comprehensive AI solution. However, internal documents reportedly indicate that Gemini’s brand awareness and user adoption metrics trail behind competitors such as OpenAI and Microsoft, reflecting a need for stronger positioning in the market.
Despite this, Gemini 2.0’s advancements in agentic AI, enabling autonomous task execution, have been met with positive reception. Its ability to handle multimodal inputs and outputs, combined with integration into Google’s ecosystem, could provide a robust offering for developers and users alike.
Read more: Google moves Gemini app team to AI research lab DeepMind
