Google has announced the launch of the latest model in the Gemini family, as well as a new coding agent for developers called Jules.
Gemini comes in two different model variants, with Flash balancing performance with speed and Pro optimizing for performance. The newest model, Gemini 2.0 Flash, is twice as fast as Gemini 1.5 Pro (first previewed in February 2024) while also achieving stronger performance.
Specifically, it features improved multimodal, text, code, video, spatial understanding, and reasoning performance across a variety of benchmarks.
Gemini 2.0 Flash will also feature new output modalities of text, images, and audio, whereas Gemini 1.5 Flash could only output text. Image and audio output is currently still listed as “coming soon” on the Gemini website, but Google says that rollout is expected next year.
Audio output is multilingual and can be spoken in eight different voices, with control over the language and accent. Image output enables users to build on previous outputs to refine generated images exactly as envisioned. In a demo Google shared, a user takes advantage of this to ask Gemini to take a picture of a car and transform the image to make the car a convertible.
Gemini 2.0 Flash can also use tools, such as Google Search, and can utilize third-party functions. “Multiple searches can be run in parallel leading to improved information retrieval by finding more relevant facts from multiple sources simultaneously and combining them for accuracy,” Shrestha Basu Mallick, group product manager for the Gemini API, and Kathy Korevec, director of product for Google Labs, wrote in a blog post.
Finally, the model can also take in streaming inputs of audio and video to enable the development of real-time, multimodal applications.
To help developers get started with Gemini 2.0 Flash, Google is releasing three starter app experiences in Google AI Studio for spatial understanding, video analysis, and Google Maps exploration.
Gemini 2.0 Flash is currently in an experimental state, with general availability expected early in 2025.
Introducing Jules, an AI-powered coding agent
The company also unveiled a new coding agent, Jules, that can handle Python and JavaScript coding tasks, such as fixing bugs.
Jules creates multi-step plans for addressing issues, can modify multiple files at once, and can prepare pull requests.
Jules is available to a selected group of testers now and will be rolled out more widely early next year.