Android Studio Rabbit 1 Canary 2 Expands Compose Screenshot Testing and Model Assignment

Google has released the second Canary build of the Android Studio Rabbit 1 cycle, and with it comes a steady expansion of the features that defined the cycle's opening salvo. Rabbit 1 Canary 2 continues the momentum around Compose screenshot testing and the flexible AI model assignment workflow, while also layering in the usual round of incremental fixes that keep a young preview moving forward.

Screenshot Testing Matures

The Compose Preview Screenshot Testing tool remains one of the headline features of the Rabbit cycle, and Canary 2 pushes it further. The tool is designed to test Compose UI and prevent regressions by generating HTML reports that let developers visually detect changes to their app's interface. It is a workflow that has historically required significant custom configuration, and bringing it into the IDE as a first-class capability is a meaningful quality-of-life improvement for teams that take visual testing seriously.

For Flutter developers who have wrestled with golden file testing, the pattern will feel familiar. Verifying that a UI remains visually consistent across refactors, platform version changes, and dependency updates is a real problem, and having an integrated tool reduces the friction of doing that verification reliably. While the tool targets Compose specifically, it reflects a broader industry push toward treating visual regression testing as a standard part of the development loop.

• Compose Previews Screenshot Testing gains further refinement in Canary 2
• HTML report generation makes visual change detection straightforward
• Model assignment lets developers pair different Gemini models to different tasks
• The new cycle continues its focus on AI-assisted development workflows

Model Assignment Takes Center Stage

One of the more interesting additions in the Rabbit cycle is the Model Assignment feature, which lets developers pick different AI models for different types of tasks inside the IDE. The idea is simple but powerful: assign a large, pro-tier model for complex reasoning tasks in Agent Mode, and a faster, lightweight model for simpler, latency-sensitive work like commit message generation or next edit prediction. This lets developers balance quality, speed, and quota usage across features.

Canary 2 builds on this foundation. The configuration lives under the Model Assignment tab in the Model Providers settings, and the tooling supports adjusting the model directly from the selector dropdown in the Agent Mode panel. When you change the model there, the Model Assignment settings for thinking features automatically update to reflect the selection, keeping the configuration coherent as you move between surfaces.

This flexibility matters because it acknowledges that not every AI task in the IDE demands the same model. Writing code is different from summarizing a diff, which is different from reasoning through a complex architectural problem. Giving developers granular control over that pairing is a thoughtful bit of design, and it positions Rabbit 1 as a cycle that treats AI not as a gimmick but as an infrastructure layer users can tune.

The Rabbit Baseline Takes Shape

The first Canary of the Rabbit cycle established the baseline by preserving the core of what came before while introducing the new AGP toolchain for the window. Canary 2 refines rather than redefines. The rhythm of the cycle is now becoming clear, and teams that downloaded the first build will recognize the direction of travel.

There is also a broader context worth noting. Android Studio has increasingly become part of the Gemini Enterprise story, and the Rabbit cycle reflects that positioning. As enterprise adoption of AI-assisted development grows, having granular model assignment and a robust local execution story becomes a differentiating capability rather than a nice-to-have.

For Flutter developers, the practical takeaway is modest but real. Canary builds remain firmly in experimentation territory, and production Flutter work should stay on stable channels. But the trajectory of the Rabbit cycle is worth tracking, because these AI and testing workflows will eventually filter into the stable default that every developer encounters.

Ultimately, Canary 2 is about consistency. It tells the story of a cycle that has a clear vision, and it reinforces the sense that Google is investing seriously in making the IDE faster, smarter, and more configurable. For developers willing to explore preview builds, it is a compelling snapshot of where the tooling is headed.