Gemini Spark vs Claude Cowork: Which AI Agent Actually Gets Work Done?

Both agents in this comparison changed their names in the autumn of 2026. Google renamed Gemini Spark to Task, so it now sits behind a Chat and Task toggle in the Gemini app. Anthropic is folding Claude Cowork into Claude itself. The two products also stopped being different shapes. Both now run tasks in the cloud while your laptop is closed, both run tasks on a schedule, both browse the web for you, and both hand back real Word, Excel, PowerPoint and PDF files. Both start at about $20 a month.

So the question is no longer which one can do the work. It is where your work lives, which country you are in, and how much model you want behind the agent. This guide explains what each agent is today and how the renames work. It then breaks down pricing, Mac and iPhone support and five real use cases, so you can match the agent to the work you actually do. If you want both models without two subscriptions, we cover that at the end.

The Key Takeaways

• Gemini Spark is now called Task (Google’s help pages say “Tasks in Gemini”). It is included with Google AI Pro ($19.99/month) and Google AI Ultra (from $99.99/month), and it is still unavailable in the EEA, the UK, Switzerland and Nigeria.
• Claude Cowork is Anthropic’s agent for knowledge work, generally available since April 9, 2026. It is merging into Claude and is included on Claude Pro ($20/month), Max ($100/month) and Ultra ($200/month).
• Both now let you pick the model. Gemini Tasks offers Flash or Gemini 4 Argon, and the strongest model in Cowork’s picker is Claude Opus 5.5. On the Artificial Analysis Intelligence Index v5.0, Opus 5.5 scores 58.4 and Argon 50.4.
• Both create Office files now, so file creation is no longer a Cowork exclusive. Cowork still wins for local folders, Microsoft 365 and coding. Gemini Tasks wins for Gmail and Google Workspace, bookings and shopping, and anyone already paying for Google AI Pro.
• If what you want is both companies’ models without two bills, Fello AI puts Claude, Gemini, ChatGPT, Grok and DeepSeek in one Mac, iPhone and iPad app, from $11.99/month.

Gemini Spark vs Claude Cowork at a Glance

Three things separate these agents now that their feature lists have converged. What they touch (Gmail and Workspace vs your folders and Office), where you can buy them (Gemini Tasks is blocked across Europe, Cowork is not), and which model does the thinking. Everything else is a consequence of those three.

The table below gives you the spec sheet up top. If you only read one section of this article, read this one and the use-case section further down.

The two products have arrived at the same product shape, a real AI agent that does the work, from different ends of the stack. Google starts in your inbox and reaches outward. Anthropic starts on your hard drive and reaches into the apps you already use. Neither is the universal answer; the right one depends on where your work actually lives. If you are still deciding whether you need an agent at all, or just generative AI or plain automation, our guide to agentic AI vs generative AI vs automation breaks down the trade-offs.

What Is Gemini Spark?

Gemini Spark is Google’s 24/7 personal AI agent, now called Task. It runs inside the Gemini app and works across Gmail, Docs, Sheets, Calendar and connected third-party apps even when your laptop is closed. At launch, Google described it as “an active partner that does real work on your behalf and under your direction.” It launched on Gemini 3.5 with the Antigravity harness and ran on Gemini 3.7 Flash from August 13. Today you choose the model yourself: Pro and Ultra subscribers can run a task on Flash or Gemini 4 Argon, and pick a thinking level.

Google explains the rename on its Tasks help page. Beta users kept saying they wanted Gemini to finish “tasks,” so “we renamed it Task to make it easier to understand.” You find it in the Chat and Task toggle next to the prompt box. Existing tasks and schedules carried over, and you find them in the sidebar.

In practice, Gemini Tasks can triage your inbox, pull data from your spreadsheets and docs, draft emails, check and reschedule calendar events, and flag new or hidden subscription fees on a credit card statement. With your confirmation, it can also act on the web. Gemini drives your local Chrome, which has to stay awake, or a remote browser in the cloud that keeps going after you close your laptop. It also connects to more than 40 apps, including Instacart, OpenTable and Adobe. Tasks keep running after you close your laptop or lock your phone. You can have up to 15 tasks running at once.

What Gemini Spark Costs

Spark started life bundled into Google AI Ultra, which Google restructured at I/O 2026 by adding an entry tier at $99.99/month and dropping the existing top tier from $249.99 to $199.99/month. Today the $99.99 tier gives 5x the usage of Google AI Pro and the $199.99 tier 20x, and Ultra starts at 20 TB of cloud storage and includes YouTube Premium. Google AI Pro at $19.99/month now includes 5 TB of storage and Gemini 4 Argon. All of it sits under Google’s compute-based limits, so a heavy task uses more of your allowance than a simple chat.

The cheaper door opened on July 23, 2026, and on July 30 Google widened AI Pro access to more than 160 additional countries. Google’s help page now says you need a Google AI Pro or Ultra subscription to assign tasks. There is no standalone price, and neither the free tier nor Google AI Plus includes Tasks. Google’s plans page still lists Tasks as an Ultra “first access” feature in select countries, but the help page is the one that describes how access works.

Geography is now the real gate, not price. Tasks are unavailable in the European Economic Area, the United Kingdom, Switzerland and Nigeria on any Google AI plan. You also have to be 18 or over, on a personal Google account. It runs only in the Gemini mobile app, the Gemini app on Mac, and the web at gemini.google.com. For the full breakdown of where each tier sits, see our Gemini pricing 2026 guide.

What Gemini Spark Can Do Today

Gemini Tasks is strongest at inbox triage, multi-source synthesis across your own Workspace, bookings through connected apps, and any job that benefits from running while you sleep. It now produces real files too. Google lists .docx, .xlsx, .pptx and .pdf among the formats a task can create, alongside native Google Docs, Sheets and Slides. It is weakest with files that live outside Google, and on Windows: local file access comes through the Gemini app on Mac, and Tasks are not part of the Gemini app for Windows.

Google calls the feature experimental and asks you to supervise it. Gemini always asks for your confirmation before it sends messages, deletes data, makes a purchase, completes a financial transaction or submits a web form. For passwords and payment details, it may ask you to take over the browser and type them yourself. Google also tells you not to schedule tasks that touch information you find sensitive, because if a schedule runs while you are offline, you may not be able to stop an unintended action.

What Is Claude Cowork?

Claude Cowork is Anthropic’s AI agent for knowledge work, with read/write access to the folders you connect and a sandboxed Linux VM for running code. It went generally available on April 9, 2026, and Anthropic documents no default model for it, only a picker that remembers your last choice. The strongest option is Claude Opus 5.5, released on September 22, with a 1-million-token context window that swallows an entire project folder, codebase or dataset. For a narrow set of flagged requests, such as offensive cybersecurity or dual-use biology, Claude re-runs the request on an older Opus model and tells you it switched.

Cowork’s defining trait is that it creates actual files, not just text in a chat window. Ask it for a marketing report and it produces a real .docx. Ask it for a financial model and it builds a real .xlsx with working formulas. Ask it for a pitch deck and it returns a styled .pptx, which you can keep editing with Claude for PowerPoint. It reaches your phone too. Cowork runs in the Claude apps for iOS and Android, so you can start a task on a Mac and check it from a phone mid-run. For the deeper walkthrough we already published, see our Claude Cowork guide.

Cowork Is Becoming Just Claude

The name is going away. Anthropic’s product page now opens with “Claude Cowork is now just Claude,” and the new Claude experience, which merges chat and Cowork into one conversation, is rolling out to Pro, Max and Ultra. If your message box still shows separate “Chat” and “Cowork” options, you do not have it yet. Team (in beta) and Enterprise plans get it only where it has been switched on, and keep Cowork as a separate mode until then. Your existing Cowork tasks, projects and files carry over.

Cowork also moved off your laptop. Anthropic’s architecture documentation states that “Cowork sessions run in the cloud by default: the agent loop and code execution run on Anthropic’s servers”, with sessions and files saved to your Claude account. Local sessions still exist on desktop, and Claude reaches your local files, its built-in browser and computer use only while the desktop app is open. Everything else keeps running when your Mac sleeps.

What Claude Cowork Costs

Cowork is included on every paid Claude plan. Claude Pro at $20/month ($17/month billed annually) gives you full Cowork access with lower usage limits. Claude Max is $100/month for five times Pro’s usage. Claude Ultra is $200/month for ten times Pro’s weekly usage, with no 5-hour limit on verified accounts. Anthropic renamed these plans on October 30, 2026; Ultra is the old Max 20x price point. Teams pay $25/seat/month for a Standard seat and $125/seat/month for a Premium seat with five times the usage ($20 and $100 billed annually). We go tier by tier in our guide to what Claude Cowork costs.

The honest catch is usage. Anthropic says longer agentic tasks that search the web, run code or create files generally use more of your limit than a quick question, and everything counts toward the same allowance. If you plan to run several long tasks a day, budget for Max.

What Claude Cowork Can Do Today

Cowork is at its best with file-heavy work that ends in a deliverable. Think file organization, research that ends in a written report, slide decks from raw notes, sales outreach and calendar planning: any task where the output is “give me the document.” It handles Microsoft 365 through a native connector, so Office users do not get penalized for being on the Anthropic side. It can also run on a schedule. Anthropic’s own example is “Every weekday at 8 AM, summarize new messages in my team’s Slack channels.” When a task needs a website, Cowork opens a browser in its side panel and fills in forms there.

Cowork’s weaknesses are narrower than they were. Scanned PDFs and complex layouts can still trip it up. Local file access, the built-in browser and computer use all need the desktop app open, and computer use is still in beta. The Linux desktop app is in beta too. On Team plans, the web and mobile versions are also in beta. Cowork activity is captured in Anthropic’s Compliance API, so the old audit-trail gap is closed.

What is no longer a weakness is the one the original version of this comparison leaned on hardest. Anthropic’s own help page puts it plainly: in cloud sessions, “Claude keeps working when you close your laptop or step away,” so background execution is not Spark’s structural advantage any more.

Which AI Agent Wins for Each Job? 5 Real Use Cases

The verdict changes by task. Here are the five use cases we get asked about most, with the honest pick for each.

Best for email and inbox triage

Winner: Gemini Spark (Tasks). This is the job Spark was built for. It lives inside your Gmail, reads incoming threads, drafts replies, scans for senders, and runs schedules triggered by events (“when my flight is delayed, notify me and propose an update to my itinerary”). Cowork can reach Gmail through a connector, but it is a second-class workflow, since Cowork was built for the file system, not the inbox.

Best for file creation

Winner: Claude Cowork, by less than it used to be. Both agents now produce real Word, Excel, PowerPoint and PDF files. Cowork keeps the edge when the inputs and outputs live in folders on your computer or in Microsoft 365. It reads and writes those folders directly, and it hands decks and spreadsheets to Claude for PowerPoint and Excel. Gemini Tasks is the better pick when the result should be a Google Doc, Sheet or Slides deck your team already shares.

Best for coding tasks

Winner: Claude Cowork. Cowork inherits Anthropic’s coding pedigree, runs code in a sandboxed VM where it can test it, and uses Opus 5.5’s 1M-token context window to hold an entire codebase in memory. Gemini Tasks can run code on a remote computer too, and Google has built its Antigravity coding agent into the Gemini Mac app. If you want a dedicated coding agent, Claude Code is still the stronger pick, but Cowork is the better general-purpose agent for engineers. Our Cowork vs Claude Code comparison breaks down which surface suits which job.

Best for shopping and bookings

Winner: Gemini Spark (Tasks). Gemini connects to more than 40 apps, including Instacart and OpenTable, and can drive Chrome to add items to a cart or book a table, stopping for your confirmation before it pays. Cowork’s built-in browser can fill in forms on any site, but it has no partner network behind it. If you want an agent that refills groceries or books dinner while you are busy, Gemini is the more practical choice, provided you live outside the EEA, the UK, Switzerland and Nigeria.

Best for research and analysis

Tie, with a tilt to Cowork. Cowork’s 1M-token context window plus its ability to write the final report as a real document gives it the edge for deep research that ends in a deliverable. Gemini Tasks wins if your source material is already in Gmail and Drive, and if you want a daily digest that compiles itself overnight. On raw model strength the gap is real but smaller than it was. On the Artificial Analysis Intelligence Index v5.0, Claude Opus 5.5 scores 58.4, ahead of Claude Sonnet 5.5 at 54.2 and Anthropic’s own Claude Fable 5.1 at 49.7. Gemini 4 Argon, the strongest model you can pick for a Gemini task, scores 50.4. Gemini’s advantage is proximity to your data, not model horsepower.

Mac and iPhone Support: Where Each Agent Actually Runs

Both agents work on a Mac, but they get there differently. Claude Cowork runs in the Claude desktop app for macOS and Windows, with a Linux version in beta. On a Mac it needs Apple Silicon (M1 or later) or an Intel Mac with virtualization support, because local sessions run in a sandboxed Linux VM. The desktop app is now one surface among several. Cowork also runs on the web at claude.ai and inside the Claude apps for iOS and Android, with the same sessions and files following your account.

Gemini Tasks run in the cloud, so the device only needs to host the Gemini app or a web browser. On Mac, you reach them through the Gemini app for Mac, launched April 15, 2026, or the web. The Mac app is also where Tasks can work with your local files, and you can send it a task from another device. On iPhone, Tasks live inside the Gemini iOS app. For how both stack up against the rest of the iPhone AI lineup, see our ranking of the best AI apps for iPhone. For the broader field of agents, see our best AI agents in 2026 list.

Which Should You Pick? The Honest Verdict

Pick Gemini Spark (Tasks) if your work lives almost entirely inside Google Workspace and you are in a country where you can actually get it. That means Google AI Pro at $19.99/month, or Ultra from $99.99/month if you want the higher limits. The strongest reasons to go are inbox automation, scheduled background work, and shopping or booking through connected apps. If you already pay for Google AI Pro for the storage or for Gemini itself, Tasks costs you nothing extra.

Pick Claude Cowork if your work ends in deliverables like reports, spreadsheets, decks or code built from files on your computer. It is also the right call if you live in Microsoft 365 or a mixed Office and Drive setup. Same if you are in the EEA, the UK, Switzerland or Nigeria, where Gemini’s country list rules you out, or if you want the strongest model behind the agent. Cowork is also the safer starting point for anyone new to agents, because it has been generally available since April and Claude Pro at $20/month gets you all of it.

Use Claude and Gemini in One App

Most people reading this comparison want two things at once: Claude for the long report and Gemini for a second opinion. You do not need two $20 subscriptions for that. Fello AI puts Claude, Gemini, ChatGPT, Grok and DeepSeek in one native Mac, iPhone and iPad app, so you can ask both and compare the answers in the same window. It can also search the web, chat with a PDF, and turn the result into a Word document, an Excel sheet or a PowerPoint deck.

Fello AI starts from $11.99 a month in most countries (prices checked October 2026), has a free tier, and holds 4.7 stars across 27,000+ reviews. It is not an agent that runs tasks while your laptop is closed, and it does not replace either one for that. It replaces the second chat subscription.

The Bottom Line on Gemini Spark vs Claude Cowork

If we had to pick one for the average Mac-using knowledge worker right now, Claude Cowork wins, but on fewer grounds than in May. It has the stronger model, it works the same on every paid plan in every country Claude serves, and it handles local files and Office without a workaround. Gemini caught up on price in July and has since added Office file output and a model picker with Gemini 4 Argon.

Still, this is a six-month story, not a two-year story. Gemini Tasks has the Workspace gravity to make 24/7 background work useful, and since it reached AI Pro it has been a default rather than a splurge. Watch the country list next, because it is the main thing still holding Gemini back, and it is European and UK readers who stay locked out.

Cowork is the agent you can rely on. Gemini Tasks is the agent most Google users already pay for.

And for everyone else, Fello AI keeps both model families one tap apart without a second subscription bill. On a Mac you can also run each side natively, using our Claude desktop client guide for macOS and the Gemini desktop client guide for macOS.

FAQ