Gemini Spark

What Spark is

Gemini Spark is a personal AI agent built into the Gemini app. Google announced it at its I/O conference on 19 May 2026.^[1] Its distinguishing feature is that it is not a chat window that waits to be asked something: it runs around the clock, 24 hours a day, and takes actions on the user's behalf under their direction.^[1] Inside the app it is reached from a Spark tab.^[1][2]

That combination changes what a language model is asked to do. An ordinary assistant finishes each turn with a reply. An agent is designed to keep a task going after the reply — planning steps, calling the tools it has been given, and reporting back. Spark's work is framed as being done under the user's direction, so the user sets the goal and the agent carries it out across the apps it can reach.

Launch

At launch Spark ran on Gemini 3.5 Flash. It first went out to trusted testers, and a beta for Google AI Ultra subscribers in the United States, aged 18 and older, was planned for the following weeks.^[1][2]

Onto the desktop

On 30 June 2026 Spark came to the Gemini macOS app.^[2] There it took on work that lives in files and folders rather than in a conversation: organising folders, creating documents from local files, and handling Google Workspace workflows. This version was offered to English-speaking Google AI Ultra subscribers.^[2]

The move is a meaningful one for an agent, because a desktop app gives it a different set of things to act on. In a chat window the agent mostly produces text; on a machine with local files and a workspace account, it can touch the documents a person actually works in.

The 3.7 Flash update

From 13 August 2026 Spark has run on Gemini 3.7 Flash, with improved tool use for Workspace apps.^[3] The same update widened who could use it: Spark became available to Google AI Pro and Google AI Ultra subscribers in more than 160 countries, though not in the European Economic Area, Nigeria, Switzerland or the UK.^[3]

Google said the model change improved accuracy and output quality for complex, multi-skill workflows in Spark — tasks that need several different tools in sequence rather than a single step.^[3] As an illustration of what that looks like in practice, Google said that with 3.7 Flash, Spark can consolidate files, draft emails and update status documents.^[3]

Where it sits

Three things stand out in Spark's short record. First, it is tied closely to Google's model releases: the agent moved from Gemini 3.5 Flash to Gemini 3.7 Flash, and each model change brought better handling of the tools it depends on.^[1][3] Second, its reach widened in stages — trusted testers, then a beta for one subscription tier in one country, then the macOS app for English speakers, then subscribers in more than 160 countries.^[1][2][3] Third, the work it is given is the ordinary paperwork of a working day: files, folders, drafts, status updates.^[2][3]

That is a useful way to read what a personal agent is for. Rather than answering a question once, it is meant to take a goal and grind through the small steps that turn a goal into a finished file or a sent draft. Whether an agent does that reliably is largely a question of tool use — how well it picks the right app, supplies the right arguments and checks the result — which is exactly what Google emphasised in the 3.7 Flash update.^[3]