At I/O 2025, Google announced a slew of updates for the Gemini app, including the launch of 2.5 Flash and Gemini Live camera and screen sharing for iOS.
Model updates
An update to Gemini 2.5 Flash, which Google first announced in April, improves its performance on reasoning, multimodality, code, and long context benchmarks, while making it more efficient by using 20-30% less tokens. It’s now available for everyone in the Gemini app, while an updated preview version (05-20) is now live in Google AI Studio and Vertex AI. It will be generally available for production to developers and enterprise customers in “early June.” Gemini 2.5 Pro will hit GA “soon after.”
Google announced Gemini 2.5 Deep Think with advanced thinking capabilities that leads across benchmarks for Mathematics (USAMO 2025), Code (LiveCodeBench v6), and Multimodality (MMMU). This “enhanced reasoning mode” leverages “new research techniques enabling the model to consider multiple hypotheses before responding.”
The company is taking “extra time to conduct more frontier safety evaluations and get further input from safety experts.” It will first be available to trusted testers via the Gemini API for feedback “before making it widely available.” Once that is complete, the Gemini 2.5 Deep Think mode will be available to consumers as part of the Google AI Ultra subscription.
We’ll continue improving Deep Think based on our frontier safety evaluations and user feedback before we release it more widely.
Google also announced Imagen 4 with ‘lifelike detail” and better text/typography outputs, as well as speed improvements. It’s available in the Gemini app today. Veo 3 with native audio generation is available in the Gemini app (US) if you have Google AI Ultra. That includes sound effects, background noises, and dialogue between characters.
New Gemini features
As for new Gemini app features at I/O 2025, Google is bringing the Project Astra-powered camera and screen sharing to the iPhone and iPad app after debuting for Android last month. The fullscreen Gemini Live interface will show new buttons to launch this. It’s available for all (free) users, and will start rolling out to iOS today.
Looking ahead, Gemini Live will get Gemini apps/Extensions support for Google Maps, Calendar, Tasks, and Keep in the “coming weeks.” From a conversation, you can have Live create Calendar events and search Maps, while you can expect support for more first-party services in the future.
Deep Research can now (starting today) combine public data with your private PDFs and images for a “holistic understanding, cross-referencing your unique knowledge with broader trends.” Gmail and Drive integrations are “coming soon.”
For instance, a market researcher can now upload internal sales figures (as PDFs) to cross-reference with public market trends, all within Deep Research. Or an academic can pull in specific, hard-to-find journal articles to enrich their literature review.
Meanwhile, when you’ve written something with Gemini Canvas, a new “Create” menu will let you generate a Web page, Infographic, and Quiz, as well as the existing Audio Overview, out of it. There’s also an open “Describe your own app” input box.
Gemini is also adding interactive quizzes:
For example, simply ask Gemini to “create a practice quiz on thermodynamics” and then dive into a tailored learning experience. As you answer, Gemini provides instant feedback, highlighting topics that need more attention. Once you’re done, Gemini proactively offers a personalized follow-up quiz, focused on the areas you found challenging, helping you turn weaknesses into strengths.
Google AI Ultra subscribers will soon get “Agent Mode” in gemini.google.com powered by Project Mariner: “Imagine simply stating your objective, and Gemini intelligently orchestrates the steps to achieve it.” You’ll have a chat at the left, while right side shows a browser window. This brings together “web browsing, in-depth research and smart integrations with your Google apps”
FTC: We use income earning auto affiliate links. More.
