Gemini Spark is Google's always-on personal AI agent: you hand it a goal, and it keeps working on it — across Gmail, Docs, Calendar, Chrome, and third-party apps — even after you close your laptop or lock your phone. As of August 13, 2026, Spark runs on Gemini 3.7 Flash, a model update that brings a large jump in multi-step task reliability. It is available to Google AI Pro and Google AI Ultra subscribers in 160+ countries — but not yet in the UK, Switzerland, Nigeria, or the European Economic Area.
At a glance
Last verified: 2026-08-22. Pricing, availability, and model versions change fast — last checked today.
- What it is: A 24/7 personal AI agent inside the Gemini app that executes multi-step tasks on your behalf, under your direction.
- Engine: Gemini 3.7 Flash (GA August 13, 2026), which moved Spark off the older model with "improved tool use for Google Workspace apps." (Google)
- Key claims: DeepSWE v1.1 jumped from 49.0% to 65.3% versus the previous model; these are Google's own numbers, not independent tests.
- Access: Google AI Pro ($20/month, US) or Google AI Ultra (from ~$100/month). Not in EEA, UK, Switzerland, or Nigeria yet.
- Verdict: The most practical "agent that does things" available to mainstream users — but treat it like a capable new hire, not a senior employee. Review its work.
What exactly is Gemini Spark?
Gemini Spark is a personal AI agent that runs continuously in the cloud — on Google-hosted virtual machines — so tasks keep moving even when your devices are off. You describe an outcome in plain language ("follow up with leads who never got a reply"), and Spark breaks it into steps, uses your connected apps, and reports back. Google announced it at I/O 2026 in May and has been expanding it since. (GIGAZINE, Google)
The distinction from a chatbot matters: a chatbot answers; an agent does. Spark is built to read your inbox, your documents, your calendar, and your task lists — and then act: drafting the reply, building the document, booking the slot.
What changed with Gemini 3.7 Flash?
On August 13, 2026, Google made Gemini 3.7 Flash generally available and switched Spark to it the same day. Google describes 3.7 Flash as its "most intelligent workhorse model," tuned for coding and agentic workflows, with better planning, more disciplined tool calls, and fewer retries when a task hits a roadblock. (Google, 9to5Google)
The headline benchmark is DeepSWE v1.1, which measures whether a model can fix real software issues end to end: 3.7 Flash scored 65.3%, up from 49.0% on Gemini 3.6 Flash — about three weeks apart. Google also reports gains on multi-step agent tasks (AutomationBench: 17.0% → 30.4%).
An honest caveat: these are Google's own published figures. No independent lab has run a same-generation head-to-head yet, so treat them as a strong directional signal and pilot Spark on your own tasks before building a process around it.
For developers, 3.7 Flash is separately available in the Gemini API at an introductory price of $0.75 per 1M input tokens and $3.75 per 1M output tokens through December 31, 2026 — a different product from Spark, but the same engine. (Google)
| Benchmark | Gemini 3.7 Flash | Gemini 3.6 Flash |
|---|---|---|
| DeepSWE v1.1 (real software fixes) | 65.3% | 49.0% |
| AutomationBench (multi-step agent tasks) | 30.4% | 17.0% |
Vendor-reported scores; independent verification pending.
What can Gemini Spark actually do for you?
Spark's value comes down to four practical capabilities, each tied to where you already work:
1. It works inside your Google apps
With your permission, Spark connects to Gmail, Docs, Sheets, Slides, Drive, Calendar, Keep, and Tasks — and can read and act across them. Recent updates let it edit private and shared Docs, Sheets, and Slides, read comments, and insert images. (CNBC-TV18, Google)
Useful pattern: "Find the emails from people who asked about my service in the last two weeks, log each one in my Client Tracker sheet, and draft a personal follow-up for each."
2. It keeps running when you stop
Because Spark executes on cloud VMs rather than your device, tasks continue after you close the laptop or lock the phone. Google frames it as "your most complex to-dos, done" — watch it work live, or let it run in the background and review the result. (gemini.google)
3. It can use Chrome the way you do
Through Chrome integration, Spark can use accounts you are already signed into and passwords saved in your browser to visit websites, compare options (flights, tools, prices), and begin bookings — while handing control back to you before sensitive actions like payments. (Digital Trends ) No new logins, no new tooling; it uses the browser you already use.
That is also the scariest sentence in this article. An agent browsing with your logged-in sessions has real power, and Google's own onboarding warns it "may do things like share your info or make purchases without asking." It is designed to request approval before high-stakes actions — but you should supervise it like you would any new hire with access to your accounts. (Decrypt)
4. It connects to third-party apps via MCP
Beyond Google Workspace, Spark gains connections through the Model Context Protocol — at launch, partners included Canva, OpenTable, and Instacart, with more planned. That turns "booking a restaurant" or "prepping a design asset" into delegable tasks, not just drafting jobs. (Decrypt)
Who can use Gemini Spark, and what does it cost?
Spark requires a paid Google AI subscription:
- Google AI Pro — $20/month (US): Spark began rolling out to Pro subscribers in the US in late July 2026, with more countries promised. (CNBC-TV18)
- Google AI Ultra — from ~$99.99/month: Spark is included for Ultra subscribers in 160+ countries where Gemini apps are supported. (Google)
Not available in the European Economic Area, the United Kingdom, Switzerland, or Nigeria. Google has not given a date for those regions. If you are in one of them, none of this applies yet. Pricing and availability change quickly — verify on Google's pages before subscribing.
How should you actually start using it?
The reliable adoption curve for any AI agent is the same: one task, reviewed, then more.
- Pick one recurring, low-risk task with a clear definition of "done" — chasing follow-ups, consolidating weekly notes into a status doc, building a first-draft comparison of three tools.
- Give Spark that one job and review every output before it leaves your hands. You are calibrating its judgment, and yours.
- Once it earns trust on task one, chain tasks. Lead intake → follow-up drafted → call booked. Chained workflows are where an always-on agent pays for a $20–$100 subscription.
- Keep humans on the edges. Let Spark do first drafts and logistics; keep final review, payments, and anything reputationally sensitive with a person.
If you want to go further than one agent, the same delegation logic scales to dedicated agents per role — see our guides on using Gemini Connected Apps to automate your work and orchestrating a team of AI agents for your business. And if you are comparing ecosystems before committing, start with ChatGPT vs Claude vs Gemini for small business — the infrastructure differences matter more than model size. For the developer-facing side of the same engine, see how to use Gemini 3.7 Flash for coding and web development.
What this means for you
If you run a business or a team, Spark is the first mainstream agent that combines three things that used to require custom automation: access to your real work apps, a browser with your logged-in sessions, and the ability to run 24/7 without your machine. The honest position is "promising early agent with a much better engine." The 3.7 Flash upgrade specifically targets the failure mode that made earlier agents frustrating — botched tool calls and tasks that silently stall. But it still makes mistakes, it still misreads intent sometimes, and it is not available everywhere. Start with one task, keep approvals on, and scale what works.
FAQ
Q: Is Gemini Spark free?
A: No. Spark requires a paid Google AI subscription: AI Pro ($20/month in the US as of mid-2026) or AI Ultra (from ~$99.99/month). There is no free tier.
Q: What is the difference between Gemini Spark and Gemini 3.7 Flash?
A: Gemini 3.7 Flash is the AI model; Spark is the always-on personal agent that runs on it. Spark switched to 3.7 Flash on August 13, 2026, gaining more reliable multi-step tool use, per Google.
Q: Can Gemini Spark access my Gmail and Google Drive?
A: Yes, with your permission. You choose which apps to connect, and Spark can then read and act across Gmail, Drive, Docs, Sheets, Calendar, Keep, and Tasks. You can watch it work or review results afterward.
Q: Is Gemini Spark available in the UK or Europe?
A: Not yet. As of August 2026, Spark is unavailable in the EEA, UK, Switzerland, and Nigeria. Google says expansion is planned but has announced no date.
Q: Can Gemini Spark spend money or make purchases on its own?
A: Google's onboarding warns Spark "may do things like share your info or make purchases without asking," while also stating it is designed to request approval before high-stakes actions. Keep supervision and approval prompts on, especially for anything involving payments.
Q: How good is Gemini Spark compared with a year ago?
A: Google's own benchmarks show a large recent jump: DeepSWE v1.1 (end-to-end software fixes) rose from 49.0% to 65.3% between the 3.6 and 3.7 Flash models, and AutomationBench roughly doubled. Independent verification is still pending, so treat these as directional.
Sources
- Introducing Gemini 3.7 Flash — Google (blog.google)
- Gemini Spark — official product page (gemini.google)
- Gemini API release notes — Google AI for Developers
- Gemini app gets 3.7 Flash — 9to5Google
- Gemini Spark rolls out to Google AI Pro in the US — CNBC-TV18
- Google launches Gemini Spark, an always-on agent at I/O 2026 — GIGAZINE
- Spark, Chrome logins and saved passwords — Digital Trends
- Gemini Spark launch analysis — Decrypt
Updates & corrections
- 2026-08-22 — First published. Availability regions, Pro/Ultra pricing, and 3.7 Flash benchmark figures verified against Google's launch materials the same day. Benchmarks labelled as vendor-reported.
Researched and drafted with AI agents; reviewed and fact-checked under human editorial oversight. How we work.
Top comments (0)