Gemini Spark: Your Phone's New AI Agent, Explained

Short answer: Gemini Spark is the persistent AI agent Google folded into the Gemini app on August 13, 2026, alongside the Gemini 3.7 Flash model that powers it. You define a task once, and Spark keeps working on it across Gmail, Drive, Docs, Calendar, Keep, and Tasks — even while your phone is off. As of August 2026, it's tied to Google AI Pro and Ultra subscriptions.

What is Gemini Spark?

Spark works differently from a normal chat assistant: instead of asking a question and waiting for an answer, you hand it an ongoing task and it keeps chasing it in the background. According to nateszerotoai's writeup on Spark, the key distinction is that Spark runs server-side on Google's infrastructure rather than on your device — so closing the app or powering off your phone doesn't stop it. When it hits a gap, it proactively asks you for the missing input instead of quietly failing or guessing.

This is Google's most concrete bet yet on an "agent layer" sitting on top of its models. We covered the model family Spark builds on in Gemini 3.6 Flash, 3.5 Flash-Lite, and Cyber; Spark is arguably the first major consumer product built on that speed-and-cost-focused lineup.

Who can access Gemini Spark?

As of late August 2026, Spark is rolling out to most Google AI Pro and Ultra subscribers, but multiple reports note it excluded the EEA, the UK, Switzerland, and Nigeria at launch. Exact regional and tier rollout details are still being clarified as of late August 2026, so treating "every Pro or Ultra subscriber everywhere" as settled fact would be premature.

For scale: the Gemini app crossed 1 billion monthly active users on August 11, 2026, just two days before Spark launched. Rolling out a persistent agent to a user base that size signals Google now treats agentic assistants as a mass-market feature, not a niche experiment.

What tasks can Spark actually do?

Spark handles concrete office chores: drafting and triaging emails, preparing meeting briefs, researching a topic across your connected files, and creating or editing documents and spreadsheets. Its native integration with Google Workspace apps means it can scan your inbox and draft replies, or pull together relevant documents ahead of a meeting without you asking for each piece separately.

Here's roughly what it can do per app:

We covered manual versions of some of this triage in Inbox Zero with AI email triage; Spark is trying to automate a slice of that workflow, though it still asks for your approval rather than acting fully unsupervised.

What's powering Spark, and what changed in Gemini 3.7 Flash?

Spark runs on Gemini 3.7 Flash, which launched August 13, 2026, about three weeks after Gemini 3.6 Flash. The model ships with a 1-million-token context window and multimodal input covering text, images, video, audio, and PDFs. According to DeepMind's model card, it shows reported gains in multi-step planning and tool use compared with the prior model.

Two of Google's published benchmark deltas:

These are reported improvements over the previous model, not absolute claims about being the best model available. Still, a 16-point jump on DeepSWE helps explain why Spark feels more reliable on multi-step chores than earlier agent attempts. On the API side, Gemini 3.7 Flash is priced at $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026, rising to $1.50/$7.50 on January 1, 2027.

For tasks that need deeper single-shot reasoning rather than task execution, Gemini 3 Deep Think remains the better fit — Spark is optimized for speed and getting chores done, not for exhaustive analysis of one hard question.

What happens to your data?

Spark needs direct access to Gmail, Drive, Calendar, and other Workspace data to do its job, which concentrates a lot of access in one always-on account. Google emphasizes minimal necessary access per task and asking for confirmation when needed, but a server-side agent that can continuously see your email and files is a genuinely new attack surface and privacy consideration for most users.

Honestly, this tradeoff is worth it for a lot of people: the time saved automating routine chores is real. But recommending an always-on cloud agent with standing inbox and Drive access to everyone, without caveats, would be irresponsible — especially if you handle sensitive business documents or financial information. Decide deliberately which tasks you hand to Spark. The separation logic in Organize AI Chats with Projects and Gems applies here too: keep sensitive work in more contained, controlled spaces.

How does Spark compare to ChatGPT and Claude?

What sets Spark apart from rivals isn't raw model quality — it's the operating model: a proactive, scheduled agent layer, while most competing features are still built around reactive chat. OpenAI's agent and Pulse-style features in ChatGPT are moving in a similar direction, toward assistants that keep working without a new prompt each time. Anthropic's Claude, meanwhile, is leaning into Cowork and a Skills ecosystem that packages tasks as modular, reusable capabilities.

Rather than ranking the three, it's more useful to match them to how you work: Spark is the lowest-friction option if your life already runs through Google Workspace; ChatGPT's agent features lean on broad web and third-party integrations; Claude suits people who want more controlled, skill-based workflows. If you want consistent behavior across assistants, see our custom instructions guide for ChatGPT, Claude, and Gemini.

First five things to try in Spark

1. Have it scan your inbox and draft replies to your highest-priority unread emails.
2. Ask it to pull a brief from relevant Drive files before your next meeting.
3. Turn a pile of scattered Keep notes into a structured Tasks list.
4. Schedule a recurring report that updates itself with fresh data each week.
5. Point it at your connected files on a specific topic and have it draft a summary document.

Start with one small, low-stakes task. It's the fastest way to see where Spark is genuinely reliable and where it still needs your sign-off.

Frequently Asked Questions

Is Gemini Spark free?

No. As of August 2026, Spark requires a Google AI Pro or Ultra subscription and isn't included with free Gemini accounts. Exact regional and tier availability details are still being finalized as of late August 2026.

Which apps does Spark work with?

Spark natively integrates with Gmail, Google Drive, Docs, Calendar, Keep, and Tasks. That lets it draft emails, prepare meeting briefs, edit documents, and build task lists directly inside those apps rather than through a separate interface.

Does Spark run on Gemini 3.7 Flash?

Yes. Spark is powered by Gemini 3.7 Flash, released August 13, 2026, with a 1-million-token context window. The model shows reported improvements over its predecessor on agentic benchmarks like DeepSWE v1.1, up from 49.0% to 65.3%.

How is Spark different from a regular Gemini chat?

A regular Gemini chat is reactive: you ask, it answers, and the conversation ends there. Spark is proactive and persistent — you define a task once, it keeps running server-side even while your device is off, and it asks you for missing information instead of stalling silently.