ChatGPT Work vs Claude Cowork vs Gemini Spark: which AI agent actually finishes the job?

Three of the biggest AI labs on earth just released, within weeks of each other, almost the exact same product. Anthropic shipped Claude Cowork. OpenAI answered with ChatGPT Work. And then Google jumped in with Gemini Spark. Same pitch every time: stop treating the AI like a chatbot you copy-paste from, and start handing it a goal so it can go do the work while you do something else.

That's a big shift, and it's worth being skeptical about. "Agentic AI" has been the buzzword of the year, but buzzwords don't finish your meeting notes or clean up your inbox. So this guide runs all three tools through the same kind of real work task - the stuff you'd actually hand off to an assistant - and breaks down where each one pulls ahead, where it falls flat, and which one deserves a spot in your actual workflow.

What are ChatGPT Work, Claude Cowork, and Gemini Spark, exactly?

Before comparing outputs, it helps to know what you're actually testing, because these three products are built quite differently under the hood.

What is Claude Cowork?

Cowork is Claude working directly with your files, folders, and apps — reading, editing, and producing real outputs on your machine, where Chat is a conversation, Cowork is a working session: you describe the task, Claude plans and executes it, and you steer along the way. You'll find it as a separate tab next to regular Chat inside the Claude desktop app, and Anthropic has been rapidly expanding where it runs.

Cowork is where you hand Claude real work. It runs on desktop, with web and mobile in beta. Claude works in the files and tools you choose and completes multi-step tasks from start to finish. You define the goal. Claude figures out how to get there. The permission model is worth understanding before you use it - in Cowork, Claude has permission to read, edit, and create files in folders you specify—so it can actually complete tasks rather than just describe how to do them.

It's also gotten more connected recently. Anthropic launched a series of connectors and plugins for Claude Cowork, and companies can now connect it to Google Drive, Gmail, DocuSign and FactSet.

What is ChatGPT Work?

On 9 July 2026, OpenAI launched ChatGPT Work — a new agentic mode inside ChatGPT that sits alongside the familiar Chat mode and the developer-focused Codex mode. You can access it directly at chatgpt.com/work, and OpenAI has published a dedicated ChatGPT Work product page with feature breakdowns.

ChatGPT Work gathers context, plans the approach, and takes action across your tools, files, and desktop apps to create polished spreadsheets, docs, and slides, powered by GPT‑5.6, bringing together context from your team's tools to turn scattered notes, drafts, and ideas into finished work. Connectivity is a real strength here - with more than 1,400 plugins available, ChatGPT can pull context from the tools and workflows you already use to help move projects forward. And it's not locked behind an enterprise paywall the way some agent products are: ChatGPT Work is available for Pro, Enterprise and Edu plans and rolling out to Plus and Business plans, and it's also available in the ChatGPT desktop app on every plan, including the free plan.

What is Gemini Spark?

Spark is the odd one out architecturally, and that's actually its biggest differentiator. At Google I/O 2026, Google announced Gemini Spark, a personal AI agent that keeps running on Google's cloud infrastructure even after you close your laptop or lock your phone. You can find it inside the Gemini app under AI Ultra plans.

Unlike Cowork and Work, which lean on your local device or session, Spark persists on dedicated Google Cloud virtual machines, continuing to work when a phone is locked or a laptop shut - Sundar Pichai described it as "your personal AI agent that helps you navigate your digital life, taking action on your behalf and under your direction." It's also the one most tightly wired into your personal data already: it monitors your Gmail, manages your Calendar, drafts documents in Google Docs, and, in the near future, will make purchases on your behalf. Third-party reach is growing fast too - it can now execute tasks in Canva, OpenTable, and Instacart via MCP.

How do you actually test these three head-to-head?

The fairest way to compare agent products isn't a synthetic benchmark - it's giving all three the exact same real task with the exact same inputs and seeing what comes back. A few tasks work especially well for this because they mirror what most knowledge workers actually need done every week:

Meeting transcript analysis. Feed each agent the same meeting transcript and ask it to extract key decisions, action items, blockers, follow-up emails, and a Slack-ready summary.
Content repurposing. Give each agent a long-form piece of content and ask it to turn it into a shorter, platform-ready version, with a filter for what actually matters versus what's just filler.
Inbox and file triage. Point each agent at a cluttered set of emails or downloads and ask it to sort, flag, and summarize what needs your attention.

Running the same prompt through all three exposes something interesting fast: each model has a distinct "personality" in how it handles ambiguity. One tends to over-explain. One tends to apply a smart filter and cut anything that doesn't serve the goal. One tends to be the most concise by default, sometimes to a fault.

What actually separates the outputs?

Here's where the differences get practical rather than theoretical.

Does conciseness or thoroughness win?

When you're extracting action items from a messy meeting transcript, verbosity is the enemy. Long, hedge-everything summaries mean you still have to read the whole thing to find what matters. The agent that applies a genuine filter - deciding what's actually worth surfacing instead of dumping everything in a bulleted list - tends to save more real time than the one that just organizes information neatly.

That filtering instinct matters even more with content repurposing tasks. An agent that asks "does this piece of content actually serve the goal of saving the reader time?" and cuts what doesn't, produces a tighter, more usable output than one that tries to force everything into a fixed template regardless of relevance.

Which agent handles file and tool access most naturally?

This is where the underlying architecture really shows. Cowork's local file permission model means it can dig into folders and documents you point it at directly, without you exporting or copy-pasting content in first. ChatGPT Work's plugin ecosystem, with more than 1,400 plugins available, means it often already has a connector for whatever tool your team already uses. Spark's advantage is different - because Google already has all your emails, it can act on Gmail and Calendar context without you feeding it anything at all.

How much oversight does each one need?

All three products are explicit that you should stay in the loop, not hand over the keys entirely. Claude Cowork lets you enable permissions settings, so Claude shows its plan and waits for your approval before anything significant. Spark takes a similarly cautious stance: Google's official Spark product page labels it "Coming soon to AI Ultra" and instructs users to "check responses, supervise closely, interrupt when needed." ChatGPT Work follows the same pattern with its Plan mode, where ChatGPT gathers context, asks questions, and creates a step-by-step plan before it starts executing.

Which one should you actually use?

There's no universal winner here, and anyone telling you otherwise hasn't tried all three on their own real work. What matters is matching the tool to your actual stack:

Pick ChatGPT Work if your team lives across a lot of different SaaS tools and you want the broadest plugin coverage with the least setup friction. It's also the most accessible price-wise since it's available in the ChatGPT desktop app on every plan, including the free plan.
Pick Claude Cowork if your work is heavily file-based - spreadsheets, documents, local folders - and you want an agent that treats your filesystem as a first-class citizen rather than an afterthought.
Pick Gemini Spark if you're already deep in Google Workspace and want an agent that runs persistently in the background, checking your Gmail and Calendar without you having to trigger it every time.

What should you do next?

Don't just read comparisons - run your own bake-off. Take one real task you did last week (a meeting recap, a messy inbox, a piece of content that needs repurposing) and feed the identical input to all three. The differences won't show up in marketing pages. They show up in the fifteen minutes you get back, or don't, from the output.

Start with whichever agent already lives inside the tools you use most, give it a real deadline, and judge it the same way you'd judge a new hire: not by what it says it can do, but by what actually lands on your desk finished.