Google’s Gemini 2.0 focuses on efficiency.

Google’s latest AI model, Gemini 2.0, prioritizes efficiency while enhancing the capabilities of its predecessors. Officially announced on Wednesday, Gemini 2.0 brings improvements over its earlier versions, particularly with the “Gemini 2.0 Flash experimental” model, which succeeds Gemini 1.5 Flash. Google’s Flash models are designed to be lightweight, optimized for tasks that don’t require the full power of a top-tier AI model, focusing instead on efficiency.

Gemini 2.0 Flash has shown significant improvements over both the 1.5 Flash and the more powerful 1.5 Pro models in various categories. It outperforms its predecessors in benchmarks like General MMLU-Pro, multiple coding tests, factuality, math, reasoning, and image and video benchmarks. Some of these improvements are substantial, such as a 7.5-point increase in the Natural2Code benchmark and a 9-point improvement in the HiddenMath benchmark. However, 1.5 Pro still outperforms 2.0 Flash in audio benchmarking (40.1% vs. 39.2%) and long context benchmarks (82.6% vs. 69.2%).

In addition to these performance boosts, Gemini 2.0 Flash introduces new multimodal capabilities, like generating AI-created images alongside text and text-to-speech audio. It can also pull data from Google Search, run code, and integrate with third-party functions.

You’ll likely encounter Gemini 2.0 Flash more frequently than you realize. Google plans to incorporate it into its AI-powered Search, specifically for generating more accurate AI Overviews. These AI summaries, which initially faced challenges, are expected to improve with Gemini 2.0, handling complex topics, multi-step queries, advanced math, multimodal questions, and even coding. Gemini 2.0 Flash is also now available in the Gemini app on both desktop and mobile web experiences.

With Gemini 2.0, Google is embracing what it calls the “agentic era,” where AI does more of the work for you. The company is pushing for greater integration of AI into its products, aiming to assist with tasks like analyzing questions, interpreting surroundings, and even completing tasks autonomously. Key updates include Project Astra, which aims to create a universal AI assistant, and Project Mariner, a Chrome extension that supports AI-assisted browsing. Google also introduced “Deep Research,” a tool that lets AI generate detailed reports based on your chosen topic or question. Once you approve its research plan, the AI scrapes the web for sources, compiles a report, and allows you to export it to Google Docs, complete with links to the original sources for verification.