Google DeepMind has introduced Gemini 2.0, an AI model that outperforms its predecessor Gemini 1.5 Pro with twice the processing speed. The model supports complex multimodal tasks and combines text, images and other inputs for advanced thinking. Gemini 2.0 is based on the JAX/XLA framework, optimized at scale and includes new features such as deep research to explore complex topics. It’s now available to developers and trusted testers and will soon be integrated into Google products like Gemini and Search.

The new model features a leap forward in speed and accuracy compared to its predecessors. For example, Gemini 2.0 Flash outperforms the previous 1.5 Pro model on key benchmarks while maintaining twice the processing speed. Additionally, it demonstrates multimodal integration by supporting tasks such as combining text and visual thinking or executing complex instructions that involve multiple types of input and output.

Source: Google Blog

Bill Jia, vice president of engineering at Google, added:

Gemini 2.0 is fully built and trained on the JAX/XLA AI framework/compiler, which we open source and share with the world. The model training took place on a large scale. Model optimization, fine-tuning, evaluation, and integration into end-user products have driven cutting-edge technology.
Today we are putting 2.0 into the hands of developers and trusted testers. And we’re working quickly to integrate it into our products, most notably Gemini and Search. Starting today, our experimental Gemini 2.0 Flash model is available to all Gemini users. We’re also introducing a new feature called Deep Research, which uses advanced reasoning and long-context features to act as a research assistant, investigate complex topics, and create reports on your behalf. It is available today in Gemini Advanced.

Gemini 2.0’s capabilities make it well suited for a number of practical applications. Highlights include:

• Project Astra, a prototype that demonstrates advanced multimodal understanding for AI assistants and can leverage Google Maps, Search and Lens.
• Project Mariner, which shows how Gemini 2.0 can perform tasks such as filling out forms or analyzing content directly in a web browser.
• Jules, a development assistant designed to integrate into GitHub workflows and help with coding tasks under human supervision.

Beyond practical tools, Gemini 2.0 finds application in the gaming field, where it can analyze gameplay in real time and provide strategic suggestions and advice. Its ability to think spatially is also being tested in robotics, and its potential applications include navigation and problem solving in the physical world.

Google DeepMind emphasizes security as a core principle in the development of Gemini 2.0. Mechanisms have been integrated to prevent unauthorized actions, protect user privacy, and combat risks such as malicious prompt injections. Additionally, the model’s design allows users to manage sensitive information through robust privacy controls.

Community feedback on Gemini 2.0 was enthusiastic. For example, Raj Nair, a CX leader, noted:

Impressive progress from Google in AI development! The capabilities of Gemini 2.0, Project Mariner, and the Coding Agent are all signs of how AI is evolving from experimental to practical applications. Integrating such advanced technology into everyday tasks, from web browsing to development workflows, will definitely transform the industry.

For more information, see the official documentation.