Claude Code is a terminal coding agent: it runs on your machine, in your repository, under your Claude subscription, and it is interactive by design. Grok Bot is a persistent cloud teammate: named Bots on one shared xAI cloud VM per account, with memory across turns, learned routines, and computer use for apps that have no API. A Bot can clone a repository and run tests, but it has no worktree model, no diff review surface, no plan gate and no model picker. Claude Code cannot keep working after you close the laptop. The practical resolution for most developers is not to choose: run Claude Code on a host that stays on, steer it from a phone, and keep Grok Bot for the office-admin lane it was actually built for.
- Different categories. Grok Bot is a general teammate that can code; Claude Code is a coding agent that is not always on.
- Grok Bot has no model picker and no dry-run mode. Claude Code has plan mode, permissions and an explicit model choice.
- Claude Code has no persistence between sessions in the Grok Bot sense, and no schedule trigger of its own.
- Bots share one cloud computer. Claude Code runs in a directory on your machine, with whatever isolation you give it.
- Cost shapes differ: Grok Bot is bundled into a Cursor or SuperGrok subscription with an uncapped meter; Claude Code rides your existing Claude plan.
- The both-together answer is a workbench, which is what Continuum is: Claude Code on your own always-on host, driven from an iPhone.
They are not the same kind of product
Search interest in this comparison is real, and the reason is obvious: both products promise an agent that does work for you, both launched into the same news cycle, and one of them has "Bot" in the name. But the comparison is closer to "assistant versus compiler" than "Chrome versus Firefox", and pretending otherwise produces a feature table that misleads in both directions.
| Grok Bot | Claude Code | |
|---|---|---|
| Category | Always-on general-purpose teammate | Terminal coding agent |
| Launched | 11 August 2026, beta | Established, iterated continuously |
| Where it runs | One shared xAI cloud VM per account | Your machine, in your working directory |
| Interface | A messaging app with named Bots | A terminal session, plus Anthropic's own surfaces |
| Persistence | Memory, files and browser sessions across turns | Session-scoped; resumable, not ambient |
| Runs while your devices are off | Yes | No, unless something else keeps the machine awake |
| Repository awareness | Whatever it clones onto the shared VM | Native. Project instructions, permissions, hooks, MCP |
| Model choice | None. Automatic routing, undocumented default | Explicit, with effort levels |
| Non-code work | Its main purpose | Not the product |
Can Grok Bot actually write code?
Yes, and the answer needs qualifying rather than dismissing. A Bot works on a persistent cloud Linux VM with a browser, a filesystem and a terminal. It can clone a repository, edit files, run a build, run tests, and push a branch. Given that the whole product was built with Cursor and sits inside the Cursor account system, this is not incidental.
What it does not have is the machinery that makes agentic coding safe and reviewable at any scale beyond one task at a time.
- No worktree model. All of your Bots share one filesystem on one computer. Two Bots working the same repository are working the same checkout. Parallel coding agents want branch-level isolation, and this design does not provide it.
- No diff review surface. You read a chat transcript, not a staged, hunk-by-hunk diff you can accept or reject.
- No plan gate. There is no dry-run mode at all: per eesel's review, a test run performs real work, navigating real sites, changing real files and calling real tools. Approval after the fact does not undo anything.
- No model control. Routing is automatic and xAI names no default model for Grok Bot anywhere. You cannot put the hard refactor on the strongest model and the mechanical rename on a cheap one.
- No cost attribution. There is an undisclosed weekly token allowance, no Grok Bot specific spend cap reported so far, and no per-repository accounting.
None of that makes Grok Bot bad at code. It makes it a general agent that can code, in the same way a competent generalist colleague can, rather than a purpose-built coding harness. For a one-off script, a small fix, or a task where the code is a means to an end, that is entirely sufficient. For a repository several people ship from, it is not the shape you want.
Where Claude Code wins
Claude Code's advantages are all consequences of the same fact: it lives in your repository on your machine, and it was designed for that from the start rather than adapted to it.
It knows the project
Project instruction files, permission rules, hooks, skills and MCP servers all attach to the repository. The agent arrives already briefed on your conventions rather than discovering them from scratch on a shared VM.
You choose the model and the effort
Explicit model selection and reasoning-effort settings mean the expensive model plans and the cheap one grinds. That is a cost lever Grok Bot structurally does not offer, because it has no picker.
Plan mode gates the writes
A plan is produced, you read it, you approve it, and only then does the agent touch files. The ordering is the opposite of a product with no dry-run, and it is the difference between reviewing intent and reviewing damage.
The repository and credentials remain on your host
The repository and credentials remain on your host; selected context is sent to the configured provider for inference. There is no shared cloud filesystem accumulating logins from every task you ever ran.
The corresponding weakness is equally structural. Claude Code is interactive. Close the terminal, or close the laptop, and the session stops. There is no ambient Claude Code that wakes up at 7am, checks a queue, and gets on with it. That is precisely the gap Grok Bot is selling into, and it is why the comparison exists at all.
Where Grok Bot wins
Four advantages, and they are real rather than concessions offered for balance.
- It keeps running with nothing of yours switched on. The VM is xAI's, not yours. For anybody without a machine that stays awake, this is the whole product.
- Computer use on apps with no API. A Bot can sign into a browser-only SaaS product and drive it. Claude Code operates a terminal, not a graphical interface belonging to somebody else's web app.
- Learn-by-demonstration routines. Walk a Bot through a multi-step path once and it persists that path as a routine it can re-run on a schedule or on demand. Claude Code has skills and slash commands, which are written rather than demonstrated.
- Persistent named identity. A Bot accumulates memory, files, browser sessions and preferences across turns. xAI's framing is that context compounds instead of resetting, and for repeated business processes that is a meaningful difference from starting fresh.
Set against that, the cost side is the one nobody puts in a comparison table. As of 22 August 2026 access requires an eligible plan and a Cursor account: SuperGrok Plus, SuperGrok Heavy, Cursor Pro+, Cursor Ultra, or Cursor Teams Standard or Premium. Cursor Ultra is $200 a month, Cursor Teams Premium is $120 per seat, and SuperGrok Plus and Heavy are reported at $100 and $300, though we could not load the xAI pricing page to confirm them. Claude Code rides the Claude plan you already have.
Two different meters
Both products cost money in ways you cannot see precisely, and the two failure modes are different enough to plan around separately.
Claude Code runs on your Claude plan, which meters agent work with a rolling five-hour window and a weekly cap operating at the same time. That is a subscription, and subscriptions fail safe: the window closes, work stops, and you find out immediately rather than on an invoice. The trap is the weekly cap, because a heavy Monday reduces what you have left on Thursday, and parallel agent work is precisely the pattern that exhausts it fastest. This is why a live gauge is worth more than a retrospective report for anybody running more than one session.
Grok Bot rides a subscription too, but with a meter behind it. There is a weekly token allowance whose size xAI has not disclosed, and per eesel's teardown there is no Grok Bot specific spend cap yet, with overage billed at raw model rates. There is also no model picker, so you cannot move cheap work onto a cheaper model. The launch-week metering display was reportedly broken outright, showing 48 percent usage as zero, and one beta tester's summary quoted in that teardown was that they had used more tokens that month than not, which is a joke with a real complaint inside it.
| Question | Claude Code | Grok Bot |
|---|---|---|
| How much of my window is left? | Visible from the CLI, or from a live gauge | A weekly allowance of undisclosed size |
| What happens when I run out? | Work stops until the window rolls | Overage at raw model rates, no product-level cap reported |
| Can I route cheap work to a cheap model? | Yes, explicitly | No. There is no picker |
| What did this repository cost? | Not from the CLI alone | Not available |
| Which subscription paid for it? | The one you were logged into | The one attached to your Cursor account |
A workbench closes the bottom three rows for coding work: Continuum shows live five-hour and weekly gauges beside the session and records spend by repository, provider, model and day, including which account paid, so the answer to "why did the weekly cap collapse on Thursday" is a table rather than a guess.
A note on the comparison pages that already exist
Before this guide, the entire "grok bot vs" modifier space contained two pages in the whole index. One compares Grok Bot with ChatGPT Work and is published by a company selling an agent product. The other compares Grok Bot with Claude Code and is published by MindStudio, which also sells an agent platform. Neither is dishonest, and vendor-written comparisons are often the best-informed writing in a young category, because the people building in it are the people paying attention.
What a vendor comparison structurally cannot do is conclude that neither of its two subjects is the answer. That conclusion is available to us and it is the honest one here: for a developer with real coding work and a preference for keeping code on their own hardware, the interesting product is neither a terminal-only agent nor a cloud teammate, but the layer that makes a terminal agent behave like an always-on one.
We are a vendor too, and the same caution applies to this page. The parts of it you should trust most are the ones sourced to xAI's own documentation, which is quoted verbatim throughout and linked at the bottom.
The answer most developers actually want
Both halves of this comparison have a missing piece, and it is the same piece. Claude Code needs somewhere to run that does not go to sleep. Grok Bot needs the review furniture and isolation that coding work requires. A workbench supplies both.
Continuum runs Claude Code, and it also runs Codex, Cursor, Grok, Gemini and OpenCode, on a Mac, Linux or Windows host you enroll. Each Code session takes its own git worktree, so several agents can work the same repository on different branches without collision. Every session exposes a plan tracker with an approval gate, an inline diff, a pull request pane and a terminal. Live five-hour and weekly quota gauges sit beside the work and spend is recorded by repository, provider and model. The desktop app, the native iPhone client and the web client all drive the same session on the same host, so closing your laptop does not stop the run.
| Setup | Always on | Repo isolation | Model choice | Review surface | Monthly cost of the tool |
|---|---|---|---|---|---|
| Claude Code alone in a terminal | No | Whatever you set up | Yes | Terminal output | $0 on top of your Claude plan |
| Grok Bot alone | Yes | No. One shared computer | No | Chat transcript | A Cursor or SuperGrok subscription |
| Claude Code in Continuum | Yes, if the host stays on | A worktree per session | Yes, and across five labs | Plan, diff, PR, terminal | $0 |
| Continuum plus Grok Bot | Yes on both | Yes for code | Yes for code | Yes for code | Free app plus your Grok Bot subscription |
The last row is not a hedge. If you already pay for Cursor or SuperGrok, Grok Bot is free at the margin and it is genuinely good at the office-admin lane. Keeping it for that and running coding sessions on your own hardware is a coherent setup, not a compromise, and neither product asks you to drop the other.
If you want the full field rather than this pairing, the alternatives guide ranks eight always-on options. If the question is really whether any of this can run on your own machine, that answer is no, with the own-hardware alternatives laid out.
Questions people ask
Is Grok Bot better than Claude Code?
They answer different questions. Grok Bot is better at persistent office work, at driving apps that have no API, and at continuing while every device you own is switched off. Claude Code is better at everything that happens inside a repository: project awareness, explicit model and effort selection, a plan gate before writes, and keeping the code on your own machine.
Should I use Grok Bot instead of a coding agent?
Only for small self-contained tasks. A Bot can clone a repository and run tests, but the harness around that is missing: no per-branch worktree isolation, no diff review surface, no plan gate and no per-repository cost accounting. Our page on Grok Bot for coding works through where the line falls.
Can Claude Code run while my laptop is closed?
Not by itself. Claude Code is interactive and stops when the session ends. It can behave as an always-on agent if it runs on a machine that stays powered on and you drive it remotely, which is what a workbench like Continuum provides: the session lives on the enrolled host and an iPhone or browser is just a window onto it.
Which model does Grok Bot use for coding?
xAI does not say, and gives you no picker. Grok 4.6 shipped on 12 August 2026 with a 500,000-token context window and is available in Grok Bot among other surfaces, so it is the plausible default, but you cannot know or control which model handled a given task.
How much does Grok Bot cost compared with Claude Code?
Grok Bot is bundled: as of 22 August 2026 you need a Cursor account plus SuperGrok Plus, SuperGrok Heavy, Cursor Pro+, Cursor Ultra, or Cursor Teams Standard or Premium. Cursor Ultra is $200 a month and Cursor Teams Premium is $120 per seat. Claude Code costs nothing beyond the Claude plan or API key you already hold.
Can I use Grok Bot and Claude Code together?
Yes, and they barely overlap. Use Grok Bot for scheduled business processes and apps with no API, and Claude Code for repository work. If you want both to keep running while you are away, run Claude Code on an always-on host through a workbench and reach it from a phone.
Does Grok Bot work with Cursor privacy mode?
No. Cursor's Legacy Privacy Mode blocks Grok Bot entirely and data storage is mandatory, per eesel's review of the product. If your team enabled privacy mode as a condition of adopting Cursor, Grok Bot is unavailable regardless of which plan you hold.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.
- Grok Bot overview (docs.x.ai) shared cloud computer, security-boundary warning, memory, routines, connectors and computer use
- Grok Bot get started (docs.x.ai) eligible plans, Cursor account requirement, platform support, Linux statement
- Introducing Grok Bot (xAI) launch date, beta status, positioning, enterprise waitlist
- Grok Bot review (eesel) compliance gaps, missing audit log, no dry-run mode, metering behaviour
- Grok 4.6 (docs.x.ai) context window, knowledge cutoff, reasoning effort, surfaces it ships on
- SpaceXAI Grok Bot coverage (VentureBeat) Cursor Teams Premium seat price, automatic model routing, SpaceXAI branding
- Continuum pricing Free, Plus, Max 100, Max 200 and Ultra plans and their weekly hosted-usage allowances