If you’ve opened Gemini and don’t see anything called “Spark,” you’re not missing a setting, it’s a separate, gated feature. Google Gemini Spark is Google’s 24/7 personal AI agent that keeps working in the background on Google’s servers, even after you close the app, to complete multi-step tasks across Gmail, Docs, and Calendar.
That’s a meaningfully different kind of product from the Gemini app you already know. Regular Gemini answers when you ask something and stops. As a genuinely agentic AI assistant, Spark keeps going on its own, on Google’s cloud infrastructure, whether your laptop is open or your phone is asleep in your pocket.
Key takeaways:
- Gemini Spark is an always-on AI agent that runs on Google’s own cloud servers, not your device, and keeps working after you close the app
- It requires a paid Google AI Pro or Google AI Ultra subscription, there’s no free version
- Both Pro and Ultra now unlock access in supported countries, Ultra just gives more compute headroom for longer tasks
- It’s excluded in the EEA, UK, Switzerland, and Nigeria, along with roughly 30 other countries as of the most recent rollout
- It connects to your data only through services you explicitly opt into (Gmail, Docs, Calendar, Drive, Photos), not by reading your screen
- Sensitive actions like payments always require your manual approval through a “Take control” step
What Is Gemini Spark?
Gemini Spark is an always-on personal AI agent, not a chatbot you open, ask something, and close. It runs on a dedicated virtual machine AI setup, part of Google Cloud, rather than on your own device, which is the specific detail that makes the “even when your phone is off” claim technically true rather than marketing language, the work genuinely isn’t happening on your hardware at all.
As always-on AI, it functions as a genuinely proactive AI agent rather than one that waits for you to open the app and ask.
You reach Spark by sending it a message, either inside the Gemini app or by emailing a designated Gmail address, part of what makes it a real Gmail AI automation tool as much as a general assistant, and it can then work independently for extended periods, checking back in only when it needs your input or approval for something sensitive.
What Model and Technology Power It
Spark runs on Gemini 3.5 Flash, the model Google introduced alongside Spark at I/O 2026, which the company positions as delivering frontier-level performance at a fraction of the cost and latency of comparable models, optimized specifically for agentic and coding-style tasks rather than general conversation.
As of this writing, 3.5 Flash remains the confirmed underlying model, though Google ships frequent updates to its Flash lineup, so it’s worth treating this as the current answer rather than a permanent one.
The agent runs on Antigravity, Google’s underlying agent harness or platform, which is the infrastructure layer connecting Spark’s reasoning to the actual tools and services it can act on, Gmail, Docs, Calendar, Chrome, and beyond, making Spark function as a genuine Google Workspace AI agent rather than a standalone bot bolted onto a search box.
What Is the Difference Between Gemini and Gemini Spark?
This is the single most common point of confusion, so here’s the direct answer: regular Gemini is a chatbot you have a conversation with, it responds to what you type and stops when you stop. Gemini Spark is an agent that acts continuously and proactively, picking up a task and continuing to work on it in the background, across multiple steps, without you needing to keep the conversation open.
A concrete example of the difference: asking regular Gemini “summarize my unread emails” gets you a one-time answer based on your inbox right now. Asking Spark to “keep me updated on anything urgent from my manager and draft a reply I can approve” sets up an ongoing task Spark continues handling on its own, checking your inbox over time and coming back to you with drafts, rather than a single answer to a single question.
When Did Gemini Spark Come Out? (Gemini Spark Release Date and Rollout)
Google announced Gemini Spark at Google I/O on May 19, 2026, initially rolling it out to a small group of trusted testers, with a wider beta for Google AI Ultra subscribers in the US following shortly after.
The rollout has expanded significantly since then, in stages worth knowing about if you’re trying to understand why a friend has access and you don’t:
- May 2026: Trusted testers, then Google AI Ultra beta in the US.
- Mid-June 2026: Ultra access expanded, though still excluding a long list of countries including Australia, Canada, India, Japan, and South Korea at that point.
- July 14, 2026: Ultra subscribers gained access in all Gemini Apps languages, wherever Gemini Apps are supported, narrowing the exclusion list down to the EEA, Nigeria, Switzerland, and the UK.
- July 16, 2026: Google AI Pro subscribers gained access for the first time, initially US-only and English-only.
- July 30, 2026: The most significant expansion, Google opened Spark to AI Pro subscribers in more than 160 additional countries, ending the US-only and English-only restriction on Pro entirely.
As of the most recent confirmed rollout status, both Google AI Pro and Google AI Ultra unlock Spark in every country where it’s currently supported, a real change from the original Ultra-only framing many people still assume is current.
Who Gets Gemini Spark? Is Gemini Spark Free?
No, Spark isn’t included in the free Gemini app. It requires a paid Google AI Pro or Google AI Ultra subscription, it’s not a feature you can unlock on the free tier no matter how you dig through settings.
A few other eligibility details worth knowing precisely, since is Gemini Spark free has a firm no as the answer: you need to be 18 or older, Google’s free AI Plus tier does not include Spark access, and work or school Google Workspace accounts are not part of this consumer rollout, this is currently a personal-account feature. You’ll also need “Keep Activity” enabled on your account, since Spark relies on that setting to function.
What differs between Pro and Ultra isn’t access itself anymore, both tiers unlock Spark, it’s compute headroom. Ultra subscribers get a larger allowance for long, complex, multi-step tasks before hitting a limit, while Pro is enough to use Spark, just with a lower ceiling on how much extended work it can do before you need to wait or upgrade.
On actual cost: Google AI Ultra currently runs around $200 a month, down from an original $250 when Spark first launched, Google adjusted the price alongside its I/O announcements.
Google AI Pro sits considerably cheaper as the mid-tier plan, commonly priced around $19.99 a month, and since Pro now unlocks Spark too, it’s the more budget-friendly way to get access if you don’t specifically need Ultra’s higher task ceiling. Confirm the exact current price directly on Google’s own subscription page, since Google has already adjusted Ultra’s price once and could do so again.
How Do I Know If I Have Gemini Spark?
Check directly inside the Gemini app rather than assuming based on your subscription alone, since country and language availability layer on top of your plan tier. Look for a Spark-specific entry point in the app’s task or agent section, if your account, country, and language combination is currently eligible, it will appear there rather than requiring a separate signup.
Is Gemini Spark Available in My Country?
Availability has expanded substantially since launch, but it’s still not global. As of the most recent confirmed update, Spark is excluded in the European Economic Area, the United Kingdom, Switzerland, and Nigeria, alongside a broader list Google describes as 33 excluded countries in total, without a full public breakdown of every one.
If you’re in one of the specifically named excluded regions, upgrading your subscription tier won’t unlock Spark, this is a country-level restriction that sits above your plan, not something a higher tier bypasses. Google hasn’t announced a firm timeline for when these regions will gain access.
The most reliable way to check your specific situation is Google’s own current Gemini Spark help page rather than a launch-date article, since this is exactly the kind of detail that’s changed multiple times already and will likely keep changing.
How Does Gemini Spark Actually Work?
A few concrete examples grounded in how Google has actually described the agent, rather than vague marketing language:
Meeting follow-up: Spark can review notes and messages scattered across your email and chats after a meeting, then draft a follow-up email summarizing decisions and action items, ready for you to review and send rather than requiring you to reconstruct the summary yourself.
Document drafting: Point Spark at a topic and relevant background material, and it can produce a working draft directly in Google Docs, continuing to refine it based on your feedback over multiple check-ins rather than a single one-shot generation.
Ongoing monitoring: Because Spark keeps running in the background, it can watch for a specific kind of update, a price change, a new email from a specific sender, a calendar conflict, and act or notify you when that condition is actually met, rather than you having to ask repeatedly.
For web-based tasks, Spark uses Chrome directly to carry out actions like filling out forms or navigating a website on your behalf, and an Android feature called Android Halo lets mobile users see what Spark is actively doing in real time, rather than treating it as a black box working invisibly.
How to Use Gemini Spark
If you’ve confirmed you’re eligible, getting started follows a simple pattern, and here’s what it actually looks like end to end.
Step 1: Confirm eligibility and enable Keep Activity. Before Spark will appear at all, your account needs “Keep Activity” turned on, found in your Google Account’s Activity Controls. This is the setting Spark relies on to maintain context across sessions, without it enabled, the Spark entry point simply won’t show up even on an eligible Pro or Ultra plan.
Step 2: Open the Gemini app and find the Spark entry point. This is distinct from a normal chat conversation, look for a dedicated Spark or agent section rather than typing into the regular chat box and expecting agent-style behavior.
Step 3: Connect the specific services you want it to use. During setup, you’ll see an opt-in menu covering Gmail, Docs, Calendar, Drive, and Photos. Connect only what the task actually needs, if you’re only using Spark for calendar management, there’s no reason to also grant it Photos access.
Step 4: Give it a task with real scope, not just a single question. Here’s a concrete example of the difference: typing “Every weekday morning, check my calendar for the day, and if I have back-to-back meetings with less than 10 minutes between them, send me a summary email flagging the conflicts before 8am” gives Spark an ongoing, scoped job with a clear trigger and a clear output.
Compare that to just asking “what’s on my calendar today,” which is really a question for regular Gemini, not a Spark task at all.
Step 5: Review and approve sensitive actions as they come up. Anything Google classifies as higher-risk, a payment, submitting a form with personal information, gets returned to you for confirmation through a “Take control” step rather than completed automatically. Read these prompts rather than reflexively approving them, especially in your first few weeks of use.
Step 6: Check in on progress through Android Halo (if you’re on Android). This feature shows you what Spark is actively doing in something closer to real time, rather than leaving you to wonder whether a background task is actually running.
Here’s what a realistic first week might look like for a small business owner testing this out: on day one, you connect only Gmail and Calendar and set up one narrow task, flagging any email from your three biggest clients for same-day review.
Through the week, you check Android Halo once or twice to confirm it’s actually catching what you expect, and adjust the task’s wording once because it was flagging newsletters from those same client domains that you didn’t actually want surfaced.
By the second week, once you trust that one task is working reliably, you might add a second one, drafting a weekly summary of unpaid invoices pulled from your email threads, rather than connecting every available service and handing over five tasks on day one.
Start narrow, in other words. Give it one specific, bounded task before handing it something broad and ongoing, the same practical advice worth applying to any new AI agent you haven’t built trust with yet.
How to Turn Off or Disconnect Gemini Spark
If you want to stop using Spark entirely, or just disconnect one specific service without shutting the whole thing down, both are handled from the same place: the Spark or agent settings section inside the Gemini app.
To disconnect a single service, Gmail or Drive, for example, without losing Spark access to everything else, revoke that specific permission from the same opt-in connections menu you used to grant it during setup.
To stop Spark from running any tasks at all, look for a pause or disable option in that same settings area, this stops background activity without requiring you to downgrade your entire Google AI Pro or Ultra subscription. If you want to remove access entirely rather than just pausing it, downgrading away from Pro or Ultra to a plan that doesn’t include Spark will remove it, though that also removes whatever else those tiers include.
Does Gemini Spark Spy on You? Is It Safe?
This is worth answering directly rather than with vague reassurance. Spark connects to your data through structured API integration, Gmail, Docs, Calendar, and similar services you explicitly grant access to, rather than reading your screen or monitoring activity outside what you’ve connected.
Google has built in specific safety mechanisms worth naming concretely rather than taking on faith: a “Take control” mode that returns higher-risk actions, payments, form submissions with personal data, to you for explicit approval rather than letting Spark complete them autonomously, and something called the Agent Payments Protocol (AP2), designed specifically to limit and control what an agent can actually spend on your behalf.
What’s honest to say plainly, and directly relevant to anyone weighing AI agent privacy tradeoffs generally: this is still a genuinely new category of product. Independent hands-on coverage, including The Verge’s review describing Spark as “shockingly good” at multi-step tasks, has also explicitly raised open questions about whether the convenience is worth the privacy tradeoff of an agent with standing access to your email and documents.
That’s a fair, unresolved question rather than something either Google’s marketing or this guide can settle definitively. The practical takeaway: review exactly what you’re connecting Spark to before doing so, rather than granting broad access by default and figuring out the implications later.
Gemini Spark vs. Claude Cowork: What’s the Difference?
These solve a similar underlying problem, this Gemini Spark AI agent landscape now includes several serious competitors, AI that does real work rather than just answering questions, from different structural starting points, and the difference is genuinely instructive rather than just naming two competing products.
Gemini Spark runs entirely on Google’s cloud infrastructure by default, always-on, working even when your own devices are off, and it’s deeply native to Google’s own apps, Gmail, Docs, Calendar, since Google controls both the agent and the services it connects to.
Claude Cowork, Anthropic’s equivalent, launched as a desktop-based agent inside the Claude Desktop app in April 2026, with file-system access, scheduled tasks, and Microsoft 365 add-ins.
It has since expanded meaningfully: as of July 2026, Cowork gained cloud execution alongside its original desktop mode, extending to web, iOS, and Android with tasks able to run decoupled from the device that started them. Most recently, as of this week, Anthropic folded Cowork directly into Claude’s main chat interface rather than keeping it as a separate mode you switch into.
The structural distinction that still holds: Spark is built around Google’s own ecosystem specifically, while Cowork leans toward desktop-level file access and Microsoft 365 integration, a better natural fit if your work already lives in Word and Excel documents on your own machine rather than natively inside Google Docs and Sheets.
Worth a brief mention here too, since it comes up in the same conversation: OpenClaw, an open-source, self-hosted personal agent project built by developer Peter Steinberger, represents the opposite bet entirely from both Spark and Cowork, running on your own hardware, a home server or even a small always-on Mac mini, rather than any company’s managed cloud. It’s the “your hardware, your data” alternative to the managed convenience both Google and Anthropic are selling.
Gemini Spark vs. ChatGPT’s Agent: What’s the Difference?
OpenAI’s equivalent is now called ChatGPT Work, launched July 9, 2026, the same week Cowork expanded to web and mobile. It sits alongside regular ChatGPT conversation as a distinct mode, rather than a separate always-on background service the way Spark is by default.
ChatGPT Work is outcome-first: you describe a finished result you want, a competitive analysis as a slide deck, for example, and it gathers information from your connected apps and files, breaks the work into steps, and continues independently for hours, then delivers a finished document, spreadsheet, presentation, or website rather than a chat reply.
The practical difference from Spark comes down to default behavior and ecosystem. Spark is built to run continuously and proactively in the background by default, deeply tied to Google’s own services. ChatGPT Work is more session-based, you kick off a defined task and it runs to completion, with a broader plugin and connected-app ecosystem spanning services beyond any one company’s own products.
Which Is Better, ChatGPT or Gemini?
Neither is a universal winner, and the honest answer depends heavily on which ecosystem you already live in rather than a raw capability comparison. If your work and personal life already run through Gmail, Google Docs, and Google Calendar, Spark’s native integration with those exact services is a real practical advantage no amount of general capability elsewhere fully offsets.
If you’re more embedded in Microsoft 365 or a broader mix of third-party apps and services, Cowork’s file-system access or ChatGPT Work’s wider connected-app ecosystem may fit your actual daily tools more naturally.
The most useful framing isn’t “which is smarter,” it’s “which one is already built around the tools I use every day,” since for an agent meant to act inside your existing digital life, that fit matters as much as the underlying model powering it.
Here’s the quick-reference version across all four agents covered in this guide:
| Gemini Spark | Claude Cowork | ChatGPT Work | OpenClaw | |
|---|---|---|---|---|
| Runs by default on | Google’s cloud (always-on) | Claude’s cloud + desktop | Session-based, on demand | Your own hardware |
| Native ecosystem | Gmail, Docs, Calendar | Claude chat, Microsoft 365 | Connected apps, files, web | Whatever you configure |
| Cost model | Included in AI Pro/Ultra | Included in Claude Pro/Max | Included in ChatGPT plans | Free, self-hosted (your hardware cost only) |
| Best fit | Already living in Google’s apps | Desktop file access, Microsoft 365 shops | Broadest connected-app ecosystem | Full control, technical comfort required |
Common Issues and Troubleshooting
You have Google AI Pro or Ultra but don’t see Spark. Check three things in order: that “Keep Activity” is enabled in your Google Account settings, that you’re 18 or older on the account, and that your current country isn’t on the exclusion list covered earlier in this guide. A subscription alone doesn’t guarantee visibility if any of those three conditions isn’t met.
Spark was working, then the entry point disappeared. This has happened during past rollout stages as Google adjusted country and language availability. If access disappears without you changing anything, it’s worth checking Google’s official Spark help page for a rollout status change before assuming something’s broken on your end.
A task isn’t running the way you expected. Revisit the exact wording of the task you gave it. Vague instructions like “help me with my inbox” give Spark far less to work with than a scoped instruction naming a specific trigger and a specific output, the same principle covered in the example prompt earlier in this guide.
You’re not sure whether a background task is actually still running. On Android, Android Halo is built specifically to answer this by showing real-time activity. Without it, check back inside the Spark section of the Gemini app directly rather than assuming silence means nothing is happening.
Frequently Asked Questions
Does Gemini Spark work on iPhone?
Spark works within the Gemini app on iOS the same way it does on Android for core functionality, though Android Halo, the real-time activity monitor, is an Android-specific feature without a direct iPhone equivalent mentioned in Google’s current documentation.
Can I use Gemini Spark without connecting Gmail?
Yes. The opt-in connections menu lets you choose which services to grant access to individually, you’re not required to connect Gmail specifically if your tasks only need Calendar or Docs, for example.
Is Gemini Spark the same as Gemini’s Deep Research feature?
No, they’re different features under the same Gemini umbrella. Deep Research is a one-time, in-depth research task that produces a report and then finishes. Spark is an ongoing, continuous agent that keeps running across multiple sessions rather than completing once and stopping.
What happens to my Spark tasks if I cancel my Pro or Ultra subscription?
Since Spark access is tied directly to an active Pro or Ultra subscription, canceling removes access to Spark along with the rest of that tier’s features. Any ongoing background tasks would stop running once the subscription actually lapses, rather than continuing on a leftover basis.
Can more than one person share access to the same Spark agent?
Based on how Google has described the feature, Spark is built around a single personal Google account rather than a shared, multi-user agent. If you need a task handled across a team, each person would need their own eligible account and their own Spark setup rather than one shared instance.
Check Your Eligibility Today
Gemini Spark represents a genuine shift from AI you ask questions to AI that keeps working on its own, but it’s still a gated, evolving product, tied to a paid subscription, a specific list of supported countries, and features Google is still actively expanding. If you’re on Google AI Pro or Ultra, the fastest next step is simply checking inside the Gemini app directly, since eligibility by country and plan has changed multiple times already since launch and is likely to keep expanding.
For more timely, accurate breakdowns of new AI agents and tools, check the blog, including our guides on Meta’s Muse AI agent and Higgsfield’s AI Motion Designer. You can also visit the homepage for more guides.
Rollout status, country availability, and pricing for Gemini Spark have changed multiple times since its May 2026 launch and are likely to keep changing. Always confirm current details directly on Google’s official Gemini Spark help page before assuming a specific figure or country status is still accurate.
