TestingCatalog spotted a button for Gemini Live embedded in pre-release builds of Google’s web app, the clearest sign yet that real-time voice conversation is moving off mobile and onto browsers and desktops. Google has offered Live on phones only since launch, one of the more conspicuous gaps against rivals that already run voice mode everywhere. Those same test builds additionally hint at a portable Skills system reaching into ordinary chats, though Google has not announced either change and both remain confined to internal test builds.

At its May I/O conference, Google said new voice features were coming to the Gemini macOS app, and TestingCatalog had already spotted voice-selection controls and a screen-sharing overlay inside that build. Live itself is still missing from the desktop app today. Testing is reportedly underway inside Google’s Trusted Tester program, which suggests Live could debut on both desktop and web in the same release instead of trickling out platform by platform.

The Skills traces point to a bigger structural shift than a new voice button. Today, Skills live only inside Gemini Spark, the autonomous-agent product locked behind Google’s AI Ultra tier, functioning there as saved instruction sets that the agent pulls in for tasks it repeats often. The new traces suggest Google wants to detach that capability from Spark entirely.

Regular chat users would reportedly be able to:

A native Skills menu and dedicated skill folders both appear to be in development, based on screenshots TestingCatalog’s X account posted on July 16.

That would put Gemini on the same footing as its two biggest rivals. Anthropic already ships Skills as a modular way to package expertise and reuse it across Claude conversations, and OpenAI runs a comparable custom-instruction and tool-packaging layer inside ChatGPT. Framed that way, Skills reads less like a feature and more like infrastructure: a portable capability layer that travels with the user instead of living inside a single product tier.

Google has already piloted lighter versions of this idea elsewhere. Chrome picked up a simplified, prompt-based skills shortcut in April, and Gemini Enterprise lets business customers call on skills mid-conversation today. Unifying those three separate implementations into one consumer-facing system would be the more significant move.

Voice matters for a different reason than Skills does. It is an interface fight, not a feature fight. An assistant confined to typing loses ground to one a user can talk to while multitasking on a desktop, terrain ChatGPT and Claude have already claimed through native voice on web and desktop apps. Pushing Gemini’s voice mode past mobile is a bet that the assistant becomes the default input layer on a workstation, not a tab someone opens only to type.

On the model side, timing stays murky. TestingCatalog flags a Gemini 3.6 Flash build already in preparation, even as Gemini 3.5 Pro keeps missing its own targets, most recently a mid-July window it failed to hit. That raises the odds a smaller stopgap model ships before either Live or Skills does. TestingCatalog’s own reporting stops short of a launch date, noting only that something from Google could land before August, with nothing confirmed and pre-release code this far out often reworked or shelved before release.

Teams standardizing on Gemini should track the Skills rollout specifically. A Spark-independent version would let free and Pro-tier users automate recurring workflows that currently require an Ultra subscription, changing the cost math for any operator paying for Ultra just to get reusable task packages.

TestingCatalog reported these findings on July 18, 2026, citing pre-release Gemini builds and screenshots its own X account posted on July 16, 2026.