Google Routes Users to Hidden Gemini 3.2 Model
A backend routing slip is delivering an unreleased model labeled Gemini 3 Fast to some users, identified as the upcoming Gemini 3.2 Flash. A visual refresh called "Liquid Glass" is rolling out across desktop and mobile.
Read the full reportQuick Summary
Reporting by NokiaPowerUser says a backend routing slip is delivering an unreleased model labeled Gemini 3 Fast to some users, which the outlet identifies as the upcoming Gemini 3.2 Flash. NokiaPowerUser also reports a visual refresh called "Liquid Glass," featuring animated gradients, floating menus, and glassmorphism, is rolling out across desktop and mobile. The site publishes a short how-to that it says can force-trigger the hidden pipeline by enabling the Canvas toggle, switching the model picker to Fast mode, and using the site generation controls. NokiaPowerUser places the leak days before Google I/O 2026.
What happened
Reporting by NokiaPowerUser documents an accidental backend routing that is delivering an unreleased model labeled Gemini 3 Fast to some users, and the outlet interprets internal metadata as the forthcoming Gemini 3.2 Flash. The same reporting describes a UI redesign called "Liquid Glass" arriving on desktop and mobile, with animated gradients, floating menus, and glassmorphism.
Technical details
Reporting by NokiaPowerUser lists a short user workflow it says can trigger the hidden routing: enable the Canvas feature, open the model picker, choose Fast mode, and optionally enable a secondary "Thinking" mode. The NokiaPowerUser post presents the feature toggle and model label as its supporting evidence.
Enable Canvas
Toggle the Canvas feature in the Gemini interface to unlock the experimental routing path.
Select Fast mode
Open the model picker and switch to Fast mode, which surfaces the unreleased backend endpoint.
Enable Thinking mode
Optionally toggle the secondary Thinking mode to access additional inference capabilities.
Trigger generation
Use the site generation controls to route through the hidden Gemini 3.2 pipeline.
Editorial analysis — technical context
Companies running multi-model fleets commonly A/B test new backends behind feature toggles and model pickers. Observed patterns in similar incidents show that server-side routing changes can expose unreleased endpoints to subsets of traffic, especially when UI toggles or feature flags are present. For practitioners, ephemeral exposure of an unreleased model can surface new performance characteristics and prompt rapid iteration on client-side UI controls and telemetry.
Context and significance
This event arrives days before Google I/O 2026, a typical timing for product rehearsals and staged rollouts. Public reporting frames the occurrence as an accidental early exposure rather than a formal release; NokiaPowerUser is the only cited outlet in the scraped material. For ML engineers and product teams, leaks like this can surface benchmarks, latencies, or behavioral differences ahead of official documentation, which complicates versioning and reproducibility for downstream evaluations.
What to watch
Reporting outlets and official channels for Google and the Gemini product line for an authoritative statement or release notes. Observers will also watch whether the routing is rolled back, whether telemetry logs appear that document differing latencies or capability tradeoffs for Gemini 3 Fast versus existing models, and how widely the UI "Liquid Glass" redesign is deployed.
Scoring rationale
The leak is notable for practitioners because it exposes an unreleased Gemini variant and a front-end redesign ahead of Google I/O, offering early empirical behavior to evaluate. It is not a landmark model release with full documentation, so the story is important but not transformative.
Fast answers to the questions people ask first
Start here if you want the model identity, the leak mechanism, or what Liquid Glass means without reading the full report.
What is Gemini 3.2 Flash?
Gemini 3.2 Flash is an unreleased Google AI model that reportedly appeared in backend routing under the label "Gemini 3 Fast." NokiaPowerUser identifies it as the forthcoming iteration of Google's Flash-class models, optimized for speed and efficiency.
How was the leak discovered?
NokiaPowerUser documented an accidental backend routing slip that delivered the unreleased model to some users. Enabling Canvas, switching to Fast mode in the model picker, and using generation controls triggers the hidden pipeline.
What is Liquid Glass?
Liquid Glass is a visual refresh spotted alongside the leak, featuring animated gradients, floating menus, and glassmorphism. It is rolling out across both desktop and mobile Gemini interfaces.
When is Google I/O 2026?
Google I/O 2026 is expected in the days following this report. The timing of the leak aligns with typical product rehearsals and staged rollouts ahead of the conference.
Why does this matter for ML practitioners?
Ephemeral exposure of an unreleased model can surface new performance characteristics, benchmarks, and latency profiles ahead of official documentation, which affects versioning and reproducibility for downstream evaluations.
Primary sources
Every claim on this page is grounded in the cited reporting or linked public discussion so you can verify the details yourself.
NokiaPowerUser
Original report documenting the backend routing slip, the Gemini 3 Fast label, and the Liquid Glass UI redesign.
iPhone in Canada
Additional coverage of the Gemini 3.2 Flash leak and its implications for Google's AI product line.
Simon Willison
Technical analysis and commentary on the model leak, feature flags, and multi-model fleet routing.
Let's Data Science
Compiled reporting with editorial analysis, scoring rationale, and contextual significance for ML practitioners.