Gemini 3.2 Flash Leaks: Models, App Redesign, and What to Expect at Google I/O 2026

The Pre‑I/O 2026 Gemini leaks, explained

In the weeks leading up to Google I/O 2026, a series of leaks began revealing possible updates to Google’s AI ecosystem. Developers and early testers spotted references to Gemini 3.2 Flash models, a redesigned Gemini mobile app, and a new agent feature called Gemini Spark.

However, most of these details come from beta builds, developer tools, and third‑party reports rather than official announcements. Google has confirmed some Gemini 3.1 releases and platform changes, but the more dramatic claims remain unverified until the conference.

Where the Gemini 3.2 leaks first appeared

The earliest clues surfaced across multiple platforms:

• iOS Gemini app builds reportedly showed a new model option labeled Gemini 3.2 Flash in the model selector.
• Google AI Studio also surfaced references to the same model family in development tools used by developers.
• Soon after, a model named gemini‑3.2‑flash‑lite‑live‑preview appeared in the Google Cloud Console model picker, suggesting a real‑time optimized variant.

These sightings matter because developer consoles and test builds often reveal upcoming features before official launches.

The rumored Gemini 3.2 Flash model family

Leaks point to a new Flash‑tier model generation designed for speed, scale, and low cost. Flash models historically focus on high‑volume tasks rather than maximum reasoning power.

Performance claims (unconfirmed)

Several reports claim Gemini 3.2 Flash could deliver:

• roughly 92% of GPT‑5.5’s coding and reasoning performance
• significantly lower inference costs

These figures appear in leak reports and secondary analysis rather than verified benchmarks, so they should be treated cautiously.

Latency expectations

One of the biggest rumored improvements is speed. Some reports say most requests could complete with latency under 200 milliseconds, which would make the model suitable for real‑time applications like voice assistants or live AI interactions.

Pricing hints

Pricing is also unclear, but leaked metadata suggests possible API costs around:

• $0.25 per million input tokens
• $2.00 per million output tokens

These numbers appear in unofficial reports and have not been confirmed by Google.

For comparison, Google’s official Gemini API documentation shows current models such as Gemini 3.1 Flash‑Lite priced around $0.25 per million input tokens and $1.50 per million output tokens in standard pricing tiers.

The Flash‑Lite‑Live variant

The model name gemini‑3.2‑flash‑lite‑live‑preview strongly suggests a version optimized for real‑time interaction. The “Live” suffix mirrors earlier Gemini Live models designed for voice‑first or streaming conversations.

The redesigned Gemini mobile app

At the same time, leaks show Google experimenting with a major Gemini app redesign.

New interface layout

Screenshots and early builds reveal several UI changes:

• a pill‑shaped prompt bar centered on the homepage
• quick access to voice input and Gemini Live
• a bottom sheet with media and file tools (Photos, Camera, uploads)
• a reorganized tools list including images, video, music, Canvas, and research features

Reports also mention animated or glowing backgrounds, creating a more dynamic interface than the earlier minimalist layout.

Unified tools panel

The redesign also appears to consolidate AI capabilities into a single panel where users can access multimodal tools such as image generation, video creation, or research workflows.

The “Liquid Glass” design influence

Some leaked screenshots describe the Gemini redesign as adopting a translucent, glass‑like interface style similar to Apple’s recent design language.

Elements reported in leaks include:

• translucent interface panels
• frosted input bars
• soft gradient backgrounds

These effects resemble Apple’s Liquid Glass design system, which emphasizes glass‑like controls and layered transparency in modern app interfaces.

However, the similarity is largely observational. There is no confirmed statement that Google intentionally adopted Apple’s design system.

Gemini Spark: Google’s rumored AI agent

Another leak points to a new feature called Gemini Spark.

Code strings discovered in a beta version of the Google app replaced the earlier name “Gemini Agent” with “Gemini Spark.”

Reports suggest Spark could function as an agentic assistant capable of performing multi‑step tasks across apps—such as organizing emails, editing documents, or completing workflows without constant prompts.

If accurate, this would represent Google’s shift toward persistent AI agents rather than simple chatbots.

What Google has actually confirmed

Despite the volume of leaks, Google has only officially confirmed a few pieces of the puzzle.

From developer documentation:

• Gemini 3.1 Flash‑Lite was released as a generally available model optimized for speed and cost efficiency.
• Google’s API documentation lists standard Gemini pricing tiers for tokens and other services.
• Deprecation notices show older Flash models being replaced by newer versions in the Gemini 3 series.

None of the official documentation confirms:

• Gemini 3.2 Flash
• Flash‑Lite‑Live models
• the Gemini app redesign
• Gemini Spark

These features remain leak‑level information until an official announcement.

What to expect at Google I/O 2026

Based on leaks and preview coverage, several announcements are widely expected:

Possible model launches

• Gemini 3.2 or other new Gemini models

Agentic AI features

• Gemini Spark or similar assistant capabilities

Android ecosystem updates

• deeper integration of Gemini across devices through the new “Gemini Intelligence” initiative

Hardware experiments

• potential Android XR hardware such as smart glasses, although this is still speculative.

The bottom line

The leaks paint a picture of Google pushing Gemini toward three major goals:

1. Cheaper, faster AI models for real‑time use
2. A more visual, modern Gemini interface on mobile
3. Agent‑style AI assistants that act on a user’s behalf

But until Google officially reveals its plans at I/O 2026, many of the most exciting details—from Gemini 3.2 Flash performance to the full Gemini app redesign—remain educated speculation based on early developer sightings and beta features.