• Google is constantly releasing new versions of Gemini AI. The latest is a major update to the 2.5 Pro model aimed at fixing problems from an earlier update. This version will become a stable release available in the Gemini app.
• The previous I/O Edition (05-06) focused on coding upgrades. The new version is better at generating code with a high score of 82.2 percent in the Aider Polyglot test, beating competitors.
• The general-purpose Gemini 2.5 Flash has left preview, but the Pro version is lagging. The last updates have attracted criticism of 2.5 Pro's performance outside of coding tasks since the 03-25 update. Google is working to address this feedback.
• The 06-05 release now supports configurable thinking budgets for developers and is expected to be a long-term stable release. Gemini 2.5 Pro will drop its "Preview" disclaimer in the coming weeks.
• Google is interested in model vibes and often relies on ratings from LMArena and WebDevArena. Gemini 2.5 debuted with a top score on LMArena and has increased its lead with the latest Gemini Pro.
• In testing, the new model makes more effort to use coherent formatting in output. It also answers a long-running test question about the color "magenta" correctly, while a 2023 version of Claude equivocates.
• Users can tinker with the next Gemini Pro in Vertex AI and AI Studio today and it will come to the Gemini app (in preview) starting today.