Google used its I/O conference to unveil Gemini 3.5 Flash and Gemini 3.5 Pro, the latest step in its flagship artificial intelligence line and a fresh push into faster, higher-quality output. The Pro version, internally known as Cappuccino, is already drawing attention for more refined creative work and improved response accuracy.
The company is pitching the new models as a meaningful upgrade over earlier Gemini versions, with output quality and speed both getting attention. They are being positioned for a wide range of uses, including generating spatially consistent game environments, optimizing UI design workflows, software development and creative content generation. In that sense, Gemini 3.5 is not being framed as a narrow release. It is being presented as a broader toolset for people building products, writing code and producing media.
That matters because the Pro variant is being described as a rival to Claude Opus 4.7 and Gemini 3.1 Pro, two models that sit near the top of the market’s current benchmark conversation. The comparison underscores how Google is treating Gemini 3.5: not as a minor refresh, but as a direct challenge to the most capable systems in the field. The models are also said to excel at complex and creative outputs, which helps explain why they are being watched closely by developers and designers.
There is, however, a catch. The same models are also described as showing occasional instruction inconsistencies and producing UI output that can become overly complex. That is the friction point in Google’s pitch. The tools appear strong enough to generate ambitious results, but not yet polished enough to remove doubt about how reliably they follow directions or how cleanly they handle interface design.
Those questions are now playing out in testing. The Gemini 3.5 models are undergoing extensive evaluation on platforms such as Alamarina and in AI battle mode environments, where different systems are compared against one another in live conditions. Early reports also indicate that some features from Gemini 3.5 are being folded into secret testing of Gemini 3.1 Pro, a sign that Google is already trying to blend the newest capabilities into its earlier lineup.
For now, Gemini 3.5 looks like a clear advance for Google’s AI effort, especially in the areas where speed, creativity and design support matter most. The more important question is whether Google can turn those gains into models that are not just impressive in testing, but dependable enough to settle into everyday use.
