Grok Bot vs Claude Cowork vs ChatGPT Work

Three products now sell the same promise: hand off a multi-step job, walk away, come back to finished work. The difference that actually decides your choice is not capability, it's the price of admission. ChatGPT Work starts on the free plan. Claude Cowork starts at $20 a month. Grok Bot starts at a tier xAI won't print a number for.

xAI launched Grok Bot on August 11, 2026, which is what pushed all three into the same conversation. Anthropic shipped Cowork in January. OpenAI's Work has been quietly sitting inside ChatGPT the whole time, and almost nobody comparing the two newer products noticed it's the cheapest way to try the idea.

Key Takeaways:

• ChatGPT Work is on every ChatGPT plan including Free, where it's limited to the desktop app. Plus at $20 unlocks it on desktop, web, and mobile.
• Claude Cowork needs a paid Claude plan, starting with Pro at $20 a month. It is not on the free tier.
• Grok Bot requires SuperGrok Heavy, Cursor Ultra, or Cursor Teams Premium. xAI's public pricing page stops at SuperGrok Plus, $100 a month, and never prints Heavy's price.
• Only Grok Bot gives each agent its own cloud computer and lets bots message each other. That's the real product difference, and it's why it costs what it costs.
• Only Grok Bot is willing to drive apps with no API and no MCP server behind them. That widens coverage and adds a failure mode the other two don't have.

---

The Access Ladder: Free, $20, or an Unpublished Top Tier

Start here, because for most people this section ends the decision.

	Cheapest access	What that gets you	Platforms
ChatGPT Work	Free, $0	Limited Work access, desktop app only	Web, macOS and Windows desktop, mobile via Remote, Codex CLI
ChatGPT Work	Plus, $20/mo	Expanded Work on desktop, web, and mobile	Same, all surfaces
Claude Cowork	Pro, $20/mo	Full Cowork, still beta on web and mobile for some plans	macOS and Windows desktop, web, iOS and Android
Grok Bot	SuperGrok Heavy	Full Bot access, early beta	macOS download plus iOS app
Grok Bot	Cursor Ultra or Teams Premium	Same, via the Cursor side of the merged company	Same

OpenAI's plan page is the one that surprised me. Free lists limited ChatGPT Work access on the desktop app, Go at $8 keeps it limited, and Plus at $20 moves it to desktop, web, and mobile. Pro starts at $100 and mostly buys usage headroom rather than new agent features.

Anthropic's ladder is simpler: Cowork rides along with Pro, Max, Team, and Enterprise, so $20 is the door, and the free plan doesn't have it at all.

xAI's ladder has a hole in it. The individual plans it publishes are Free at $0, SuperGrok at $30, and SuperGrok Plus at $100. Grok Bot appears on none of those. It lives on SuperGrok Heavy, which shows up in the comparison table as a column with a button and no figure, plus the two Cursor tiers that came over when the companies merged. Community threads have discussed Heavy at roughly $300 a month with promotions near $99, which is consistent enough to plan around and not solid enough to quote as fact. I dug into that gap separately in what SuperGrok Heavy actually unlocks.

---

What Each One Actually Runs On

The phrase "its own computer" is doing real work in xAI's marketing, and it isn't just marketing.

Grok Bot gives each bot a machine in the cloud. Bots sign into your tools the way a colleague would, keep going when your laptop is shut, and, this is the part nobody else offers, message each other. You can put several in a group chat where they hand work between themselves, assign ownership, and only pull you in for a judgment call. If your actual problem is that you are the bottleneck between four half-finished workstreams, that architecture is the reason to look at Grok Bot rather than the two cheaper options.

Claude Cowork runs cloud sessions that follow you between devices, with direct local file access when you're on the desktop app and browser work handled through Claude in Chrome. Anthropic's docs say long-running tasks avoid timeouts and that tasks can be scheduled or recurring. What Cowork does not do is share a session between two people, and live artifacts plus local MCP servers stay desktop-only.

ChatGPT Work is a task surface inside ChatGPT rather than a separate machine. It pulls context from your files and plugins, breaks a goal into steps, works for hours while you watch, and returns documents, spreadsheets, decks, and Sites you can review. It runs on GPT-5.6, which OpenAI ships in three variants: Sol for the hard problems, Terra for balance, Luna for speed. Scheduled tasks are included on the paid plans.

One consequence worth planning around. A task surface inside a chat app inherits the chat app's permissions, which is convenient. A bot with its own computer and its own logins is closer to a service account with a keyboard, and it needs to be governed like one.

---

Memory and Routines: Three Different Promises

This is where the marketing diverges most and where the documentation is most useful.

xAI claims the fullest version. Bots remember conversations, pick up how you like things done, and, the interesting one, learn a routine by watching you do the job once, then run it themselves next time. If that survives contact with a redesigned form field, it is the most valuable feature any of these three has shipped. Nobody outside xAI has tested it yet.

Anthropic documents a limit its competitors don't mention: memory does not transfer from an ordinary Claude chat into a Cowork session unless you're inside a project. Months of teaching Claude your preferences in conversation do not automatically arrive when Cowork starts working. Cowork also spends your usage allowance faster than chatting does, which Anthropic says outright.

OpenAI's version leans on the plan ladder instead. Memory, memory with past chats, and memory sources all scale by tier, and Work inherits whatever the account has. Scheduled tasks and projects carry the continuity.

The distinction that matters across all three is one we've written about at length: stored preferences are not the same thing as a memory of work in progress. The difference between agent memory and chat history is the thing to test in week one, not month three.

---

Where Each One Breaks

Nobody's launch page leads with this, so here it is in one place.

Grok Bot is labelled early beta by xAI itself, ships on macOS and iOS as far as xAI's own page shows, and makes enterprise buyers join a waitlist with no date. Every productivity figure in the announcement is attributed to xAI staff, including a salesperson who says he trusts a bot to run without reviewing its output. Some outlets reported Windows and Linux builds on launch day; I could not find either on xAI's page. And driving interfaces instead of APIs means an app redesign can silently break a routine that an API call would have survived.

Claude Cowork's seams are documented rather than discovered: no session sharing between people, memory that doesn't cross over from chat, desktop-only live artifacts and local MCP servers, and a faster burn against your usage limits. Web and mobile are still beta on several plans.

ChatGPT Work's constraint is the plan matrix. The word "Limited" does a lot of quiet work on the Free and Go tiers, and model access differs by plan in ways that change what a long task can actually finish. If a plugin gets disconnected, tasks depending on its live data fail even when the conversation context survives.

The pattern behind all three is old. When Gartner predicted in mid-2025 that more than 40% of agentic AI projects would be cancelled by 2027, the reason wasn't weak models. It was autonomy granted before anyone built the scaffolding to catch a bad run.

---

Grok Bot vs Claude Cowork: Picking Between the Two New Ones

If you want one sentence: pay for Grok Bot when you need several agents working in parallel and coordinating without you, and pick Cowork when you want one capable agent on a plan you're probably already paying for.

The tie-breakers, in the order I'd apply them. Are you already on Cursor Ultra? Then check the Bot app before you buy anything, because you may already have access. Is your machine a Windows machine? Then xAI's own download page is a problem and Cowork covers Windows today. Does the work touch money, customers, or anything you can't quietly undo? Early beta plus unverified autonomy is a bad combination this month regardless of budget.

And if the honest answer is that you have one recurring job that needs to run at 6am, neither of these is the interesting question. You're paying for an org chart you don't have.

MoClaw runs that shape of work on a hosted cloud computer for $20 a month with 1,000 credits, and there's a three day trial with 1,000 credits if you want to test a real workflow before committing. It's a managed take on OpenClaw with scheduled tasks, persistent storage, browser control, and delivery into Slack, Telegram, or email. No bot-to-bot group chat, and no unpublished number at checkout. Start the trial if you'd rather test than read.

---

ChatGPT Work vs Claude Cowork for Everyday Deliverables

For the boring, valuable work, documents that need to exist by Thursday, these two are closer than the branding suggests, and the choice usually comes down to where your files already live.

Cowork's strength is local. On desktop it reads your actual filesystem, and browser work runs through Claude in Chrome, so a task that spans a folder and six tabs feels natural. Work's strength is the plugin surface and the output formats: documents, spreadsheets, presentations, and Sites, produced in a review-ready state, with approvals requested before consequential actions.

Cost is the tiebreaker for most people. Both sit at $20 a month at the useful tier, but Work also exists in limited form on Free and on the $8 Go plan, which makes it the cheapest way to find out whether you even like delegating this way. Try it there for a week before you pay anyone. Our guide to what ChatGPT Work is covers the task surface in more depth.

---

FAQ

Grok Bot vs Claude Cowork: which is better?

Grok Bot is better when you need multiple agents coordinating in parallel, each with its own cloud computer, and you can absorb early-beta risk on a top-tier subscription. Claude Cowork is better for one agent doing capable work on a plan starting at $20 a month, on both macOS and Windows, with the limits documented up front.

Is ChatGPT Work free?

Partly. OpenAI's plan page lists limited ChatGPT Work access on the Free plan, restricted to the desktop app. The $8 Go plan keeps it limited, and Plus at $20 a month expands it to desktop, web, and mobile.

What plan do you need for Grok Bot?

SuperGrok Heavy, Cursor Ultra, or Cursor Teams Premium as of August 12, 2026. SuperGrok at $30 and SuperGrok Plus at $100 do not include it, and enterprise access is a waitlist.

Which one works on Windows?

Claude Cowork and ChatGPT Work both list Windows desktop support. xAI's Grok Bot page offers a macOS download with an iOS app, so verify Windows availability before subscribing.

Do any of them need MCP or API access to use my tools?

Grok Bot explicitly does not: xAI says bots sign in and operate apps and websites directly, including ones with no clean API or MCP server. Cowork uses Claude in Chrome for browser work and supports local MCP servers on desktop. Work pulls from files and plugins.

Can bots hand work to each other?

Only in Grok Bot today. xAI supports bots messaging each other and coordinating in group chats where they pass work and assign ownership. Cowork and Work both run a single agent per task, with parallel workstreams inside one session rather than agents negotiating between themselves.

---

Pick by the Access Ladder, Not the Demo Video

Every one of these three demos well. What separates them in month three is whether you can afford to keep the subscription that hosts the workflow you came to depend on, and whether the vendor told you the truth about the limits before you built on them.

On that second test, Anthropic's documentation is currently the most honest, OpenAI's plan matrix is the most legible, and xAI's launch post is the most exciting and the least verified. On the first test, start where it costs nothing, prove the workflow, then pay for the tier that fits the workflow you actually built.

Fact-check, August 12, 2026: ChatGPT plan prices and Work availability read off chatgpt.com/pricing and OpenAI's learn.chatgpt.com feature documentation. Claude plan prices and Cowork behaviour from claude.com/pricing and Anthropic's "Get started with Claude Cowork" help article. Grok Bot tiers, platforms, and capabilities from x.ai/news/introducing-grok-bot and x.ai/pricing. SuperGrok Heavy's price is not published by xAI, so the $300 and $99 figures are reported as community discussion only. No independent benchmark of any of the three had been published at the time of writing.