Google launches Gemini 2.0 AI model that can 'think ahead'

Google has launched Gemini 2.0, its most advanced generative artificial intelligence model yet, marking a shift from information retrieval to action-oriented "agentic" AI – systems that can think, plan, and take action on users' behalf. Sundar Pichai, chief executive of Google’s parent firm Alphabet, said Gemini 2.0 is the company’s “ most capable [AI] model yet” and brings the company closer to its aim of building a “universal assistant”. Unlike its predecessor, Gemini 1.0, which focused on organising and understanding information, Gemini 2.0 introduces agentic AI. Agentic AIs are systems that can take initiative, make decisions, and perform actions on behalf of users, guided by human input and supervision. An agentic AI could automatically book a hotel, suggest activities, make dinner reservations, and provide a schedule – all based on users’ past preferences. “Over the last year, we have been investing in developing more agentic models, meaning they can understand more about the world around you, think multiple steps ahead, and take action on your behalf, with your supervision,” Mr Pichai said. The approach allows AI to make decisions, execute multi-step instructions and use tools like Google Search, Lens, and Maps. It can generate multimodal outputs, such as text, images, and audio, and integrate with the company’s other products like Project Astra, a universal AI assistant, and Project Mariner, a browser-based task manager, Google said. Millions of developers are now using Gemini, helping the company to improve its products – Search, YouTube, Android, Chrome, Google Play, Gmail and Google Maps, which are together used by about 2 billion users globally, Mr Pichai added. Previously known as Bard, Gemini is designed to support users in various tasks, from conversational interactions to creative projects. It leverages advanced machine learning and natural language processing technologies, making it a direct competitor to platforms such as OpenAI’s ChatGPT and Microsoft’s Copilot. This month also marks one year since Google launched its Gemini 1.0 AI model in December in a direct response to Microsoft-backed ChatGPT. On Wednesday, Google also launched a feature called Deep Research, which uses advanced reasoning and long context capabilities to act as a research assistant. It will be useful in handling complex topics and compiling long reports on users’ behalf, Mr Pichai said. Following the announcement, Alphabet's share price jumped more than 5.46 per cent to close at $196.71 on Wednesday, giving the company a market capitalisation of $2.53 trillion. Gemini 2.0 Flash, a testable version of Gemini 2, was made available to developers a few weeks ago for building and testing apps and to collect initial feedback. From Wednesday, Gemini users worldwide can access a chat-friendly version of 2.0 Flash by selecting it from the model menu on desktop and mobile browsers. It will also be available in the Gemini mobile app soon. Demis Hassabis, chief executive of Google DeepMind, an AI research lab that is building systems that can reason, learn, and solve complex problems, said the company will expand Gemini 2.0 to more Google products next year. “Our goal is to get our models into people’s hands safely and quickly,” he added. As a next step, the company will integrate Gemini 2.0 into Google AI Overviews to tackle more complex topics and multi-step questions, including advanced maths equations and coding. It started limited testing this week and will be rolling it out more broadly early next year. Introduced in May, AI Overviews are AI-generated summaries that appear at the top of search results, providing additional information on complex queries. Google said it is “mindful of privacy” while designing its AI agents. “We prioritise protecting users' data … this includes leveraging industry-leading security infrastructure. We ensure users have full transparency, allowing them [users] to see what data is being collected, delete it if they choose,” Dawn Bloxwich, director of responsible development and innovation at Google DeepMind. Ms Bloxwich said users can also monitor if any third-party apps are accessing their information and they can opt out. “We recognise … the many questions AI agents open up for safety and security. That is why we are … conducting research on multiple prototypes, iteratively implementing safety training, working with trusted testers and external experts and performing extensive risk assessments and safety and assurance evaluations, Mr Hassabis said. Google said it is working with its responsibility and safety committee, an internal review group, to identify and understand potential risks. Gemini 2.0's reasoning capabilities have enabled it to detect risks and automatically generate evaluations and training data to mitigate them, the company said. Despite significant investments, Gemini has yet to show any significant contribution to the company's sales. Google services business accounted for about 86.6 per cent of the company’s total sales in the third quarter of this year. It added about $76.5 billion to overall revenue, about 12.5 per cent more on an annual basis. The company did not disclose the contribution of Gemini. To quantify the contribution, in October Google announced that the Gemini app team, currently within the Google Services division, will join the AI research lab DeepMind. In October, Google also announced the creation of the AI Opportunity Initiative for the Middle East and North Africa region, through which it will contribute $15 million until the end of 2027 to help AI opportunities reach more industries and a wider range of users. However, industry analysts said there is a scope in the long-run given the size of the industry. The global generative AI market size, estimated at $16.87 billion this year, is expected to grow at a compound annual growth rate of 37.6 per cent from 2025 to 2030, according to Grand View Research.

Google launches Gemini 2.0 AI model that can 'think ahead'

Latest model introduces 'agentic AI' that can make decisions and act on behalf of users