Gemini 2 is the latest AI model from Google and the first version of the family to launch is the fast, yet powerful Gemini 2 Flash. The company says this is the start of its 'agent era' where AI is capable of performing tasks without human input.
Google launched its Gemini AI just over a year ago, bringing in a new era for the search giant that saw the rollout of AI overviews, the Gemini chatbot and more.
Initially only available to developers or as an experimental model for Gemini Advanced subscribers, Gemini Flash 2 outperforms the previous version on almost all benchmarks despite being smaller and faster.
CEO Sundar Pichai says Gemini 2 has advanced reasoning capabilities and these will also be coming to AI Overviews in Google Search to offer more accurate responses to complex, multi-step questions.
Gemini 2 will likely also have a Pro version, with the model powering all of the Google Gemini products including the Android app, chatbot and experiments.
What's new in Gemini 2?
Gemini 2 is being dubbed the 'agent era' by Google. It is a model capable of advanced reasoning similar to OpenAI's o1 but can also natively output images, speech, text and more.
The first model in the family is Gemini 2.0 Flash but the current release is branded as 'experimental'. Google says it is twice the speed of Gemini Pro 1.5, the previous flagship model, while also outperforming it on key benchmarks.
Demis Hassabis, CEO of Google DeepMind describes Gemini 2.0 Flash as the "workhorse model" with low latency and enhanced performance. It can natively generate images, text and speech where previously Gemini had to call on other models like Imagen to perform those tasks.
It also outperforms all previous Gemini models at reasoning, is significantly better at visual understanding, can translate speech at speed from audio and can analyze video better than Pro 1.5. Video analysis was already a Gemini special feature.
What about agents in Gemini 2.0?
When Gemini 1.0 launched we were in the 'chatbot' era of AI models where you could converse with them and use them to create content. Then, with the arrival of OpenAI o1 we entered the reasoning era, and simultaneously the agent era.
Agents, in AI terms, are where a model can create versions of itself to perform a range of functions on your behalf. Google has also launched a new agent tool in Gemini that can go off and browse the web for you, and return a report on a complex topic — this is known as Deep Research and is built into Gemini Advanced.
Hassabis explained: "The practical application of AI agents is a research area full of exciting possibilities. We’re exploring this new frontier with a series of prototypes that can help people accomplish tasks and get things done."
He added that this includes Project Astra — a universal virtual assistant revealed during Google I/O and the new Project Mariner. This "explores the future of human-agent interaction, starting with your browser," he said, as well as Jules which is a code agent to help developers.
Where can I try Gemini 2.0 Flash?
Gemini 2.0 Flash is currently available to subscribers of Gemini Advanced in the drop down model menu. It is labelled as experimental but I found it works fine.
You can also use it as a developer in the Gemini API or the powerful Google Gemini AI Studio. Google says it will be coming to all Google products in 2025.
More from Tom's Guide
- OpenAI announces '12 days of OpenAI' with a new product or model every day
- I just saw the future of gaming — Google's Genie 2 can turn text into a playable game in real-time
- ElevenLabs drops new conversational AI — it’s as natural as chatting to a human
