2026-08-26 · Guide
Grok Bot Free Trial: How the Metered Credit Actually Works
The Grok Bot free trial, announced with the 21 August 2026 access expansion, is the first time people without an eligible paid plan can touch the product. It is also the easiest way to learn the wrong lesson. The trial is limited usage. It is not a week of unrestricted agents. One long assignment, especially a browser crawl, can consume most of the credit. Spend it like a lab: one named bot, one reversible job, a boundary that forbids send and pay, and a written note of what finished.
Do not copy launch-week posts that still describe a Mac-only Ultra product with no trial. Supported clients are macOS, Windows, and iPhone on iOS 18+. Linux desktop, Android, and iPad are not. Paid doors after the trial start at Cursor Pro+ at $60 a month for individuals and Cursor Teams Standard at $40 a seat. The trial itself has no published numeric credit on the SAFE list, so this page will not invent one.
Treat the trial as a compute sampler, not a holiday
SpaceXAI described the expansion as limited usage. Cursor's Grok Bot billing material, where it exists, frames a trial as usage with a window, consumed by agent steps and tokens. That is closer to a sampler of the cloud computer than to "seven days of SuperGrok Heavy." Plan the work as if the credit is small. If you finish the job with credit left, you learned the product. If you burn it on a meandering browse, you learned that browsing is expensive.
| Trial behaviour | What it teaches | What it wastes |
|---|---|---|
| One research brief with sources | Whether the VM, browser, and citations work | Almost nothing if you stop when blocked |
| Connect Gmail, Slack, bank, CRM | How plugins feel | The credit, plus a shared session mess |
| Ten bots, group chat, routines | The demo | The trial, and any hope of a clean read |
| Overnight crawl of the public web | Token burn | The credit, often before morning |
The setup guide is how you install. This page is how you avoid emptying the trial on the install.
Confirm you are actually on the trial path
Eligible paid plans already include Grok Bot. If you are on Pro+, Ultra, Teams, SuperGrok Plus, or Heavy, you are not "trying Grok Bot." You are using the product you already bought. The trial is for everyone else. SuperGrok $30 and Cursor Pro $20 still do not include Grok Bot as a plan feature. The trial is the way those accounts sample it without an immediate upgrade, with whatever limits the vendor attached that day.
Download from the official bot surface, sign in, and check which entitlement the app shows. Do not assume a ChatGPT-style free tier. There is no documented standalone free Grok Bot plan. Privacy Mode (Legacy) blocks Grok Bot entirely, so a trial will not save you if that mode is on.
Pick a job that cannot send, pay, or publish
The trial is the worst moment to discover that an approval does not reverse completed work. Keep the job inside a draft. Public-source research is the cleanest. Inbox drafts that never send are acceptable only if you understand that signing into Gmail puts a session on a computer shared with every bot on the account. Prefer not to sign into mail on a trial VM you may abandon.
Paste this and do not expand it:
Name: Trial Researcher
Job: One question, public sources, one brief
You answer one question I name, using public pages only. Every material
claim has a URL. If you cannot open a source, say blocked and stop.
Boundary: Never send email, never post, never purchase, never create
accounts, never sign into sites, never use cookies from other bots.
Do not continue browsing after the brief is done.
Deliver: one page plus a source table. No appendix of extra links.
Lead Scout is the longer catalog version of that job. Inbox Triage can wait until you are on a paid plan you intend to keep, because mail sessions persist on the shared computer after the trial mood has passed.
Do not connect plugins until the first brief succeeds
Plugins are Settings, then Plugins, then Add, then a browser auth. On a trial, each connection is a credential on a VM you might not keep. Hosted MCP tokens stay with Cursor's backend. Browser sessions stay on the computer and survive bot deletion. The trial is the wrong time to find that out with a bank login.
If the first brief is useful without plugins, you have your answer: the computer, the browser, and the model-as-served can do research. If it is not useful, plugins will not fix a bad brief. They will only widen the blast radius.
Schedule nothing until you have watched a run finish
Routines assign a workflow to one bot, with a maximum of 50 per bot and 20 recent run records kept. Deleting the bot deletes the routines. A trial user who schedules a five-minute job around the clock will not get a week of automation. They will get a burned credit and a bot they then delete, taking the routine with it. Run once, watch, correct, maybe run once more. Save the method in a document you own, not only inside the product.
Teach-by-demonstration records up to ten minutes of visible computer interaction, browser workflows only, no microphone, draft skill, unavailable on iPhone. That is a poor use of a metered trial unless the workflow is short and you will keep the account.
Expect Windows and iPhone, and stop expecting Linux
Stale trial writeups still say Mac only. The FAQ lists macOS (Apple silicon and Intel), Windows (x64 and Arm64), and iPhone on iOS 18+. It also answers the Linux desktop question with no. Android and iPad are not supported. If your only machine is a Linux workstation, the trial will not give you a client. The agent computer is a managed Linux VM in the cloud, which does not install a desktop app on your ThinkPad.
iPhone can pause and resume. Editing, history, testing, and deleting need desktop. Starting a trial exclusively from a phone is how you cannot inspect what you just burned.
Budget the trial against a paid door you would actually buy
When the credit ends, the next step is a plan, not another secret free tier. Cursor Pro+ at $60 is the cheapest documented individual paid path. SuperGrok Plus is $100. Teams Standard is $40 a seat. Ultra is $200. Do not use unpublished Heavy prices. Grok Bot cost explains the weekly allowance plus on-demand overflow you will live with after the trial. There is still no Grok Bot-specific spend cap.
| After the trial | Individual | Team |
|---|---|---|
| Cheapest documented paid door | Cursor Pro+ $60/mo | Teams Standard $40/user/mo |
| Also eligible | SuperGrok Plus $100, Ultra $200, Heavy (price unpublished here) | Teams Premium $120/user/mo |
| Still not included | Cursor Pro $20, SuperGrok $30, Hobby | n/a |
If the trial job was not worth $60 a month of recovered time, do not upgrade "to keep tinkering." Tinkering is how overflow billing starts.
Recognise the five ways a trial fails
| Symptom | Cause | Fix |
|---|---|---|
| Credit gone after one evening | A long unattended browse | One question, stop when blocked |
| Nothing useful, lots of prose | No output contract | Demand a table and URLs |
| Gmail still signed in after you deleted the bot | Shared computer, deletion does not clean sessions | Revoke in Google, not only in the bot list |
| iPhone trial you cannot inspect | Mobile cannot edit or show history fully | Use desktop for the trial |
| "It sent something" | Brief without a send boundary | Kill the connection, treat as a failed test |
The last row is not a cute lesson. An approval does not reverse completed work. A trial that sends is a trial you should end.
Keep the human in the only loop that matters
The trial exists to show you a packet: a brief, a draft, a table. It does not exist to replace judgement. If you cannot spare twenty minutes to read the output, you will either rubber-stamp it or ignore it. Both waste the credit. Chief of Staff is a briefing pattern, not an autopilot. Copy that restraint.
Write down what you will not reconnect
If you abandon the trial, revoke OAuth grants at the provider. Do not trust bot deletion. Do not trust "I only used it for research" if you signed into anything. The static egress IPs on the VM can also look like a datacenter to some services. A trial login from a flagged IP is another reason to keep the job on public pages.
Verify the trial taught you something
You should leave the trial able to answer:
- Did I get one artifact I would reuse?
- Did I stay inside the boundary?
- Do I know the paid door I would use next, with a real price from the vendor?
- Did I leave credentials on the VM that I have now revoked?
If (1) is no, do not upgrade. If (2) is no, do not upgrade. If (3) is "whatever the YouTube video said," read Cursor pricing and xAI pricing yourself. If (4) is unchecked, the trial is still open in the worst way.
Keep a written log of every click during the trial
The trial is short. Memory is not a source. Write: time started, brief used, plugins added, what the packet contained, what you revoked. If you upgrade later, that log is the onboarding doc. If you do not, it is the record that you did not leave a bank session on a VM. Teams that skip the log repeat the same connections in week two and call it a new test.
| Field | Why it exists |
|---|---|
| Start time and client (Mac, Windows, iPhone) | Mobile-only trials hide history |
| Exact brief | You will misremember the boundary |
| Plugins and OAuth grants | Deletion does not revoke them |
| Artifact produced | Worth is an artifact, not a feeling |
| Revoke checklist signed | The trial is not over until this is done |
Do not use the trial to evaluate group chat or a roster
Group chat and multi-bot coordination are the demo. They are also how you share one Gmail cookie across workers before you understand the computer. A trial that starts with three bots is not a Grok Bot trial. It is a credential-spread trial. One bot. One brief. Then stop. The paid product is where you can afford to be wrong about a second worker. The trial is not.
Compare the trial against a single Cowork or chat afternoon
If a ChatGPT or Claude session on a $20 plan produces the same brief, with sources, in an afternoon, the trial just told you that you did not need a cloud computer for that job. That is a successful trial. It is not a failed product. Grok Bot is for standing work that outlasts the lid. If your work is a conversation, keep the conversation. Upgrade only when the shut-lid test is the requirement.
Write the upgrade rule before the credit hits zero
Decide now: I upgrade only if I used the artifact in a real meeting or a real review, and I have revoked every unused login. If you wait until the paywall, you will upgrade to avoid losing a toy. That is how overflow billing starts next week.
Rehearse the revoke steps while you still remember the passwords
Google, Microsoft, Slack, GitHub, Notion: each has an applications or sessions page. Open them before you start the trial and keep the tabs. When the trial ends, you will not want to hunt. A trial that only used public web still needs a check, because the bot may have tried a login you did not notice in the preview. Watch the Agent Computer once. If you never opened it, you never saw what it typed.
| Provider | What to revoke | What people forget |
|---|---|---|
| Third-party access, browser sessions | I deleted the bot | |
| Slack | App installs, user tokens | Workspace-level install |
| GitHub | OAuth apps, personal tokens | Fine-grained tokens in settings |
| Notion | Connected apps | Integration still listed |
Use a dedicated browser profile if you must log in at all
A trial VM that inherits your daily profile, via a confusing plugin flow or a copied session, is how personal mail and work mail collapse. Prefer public sources. If you must authenticate, use an account created for the trial, with no production data, and kill it afterward. Dedicated accounts are cheaper than a shared-computer incident.
Stop when the packet exists, even if credit remains
Leftover credit is not a reason to connect Slack since it is already paid. Leftover credit is a successful sampler. The product will still be there on an eligible plan tomorrow. The session you create tonight will still be there too, which is the problem. End the lab when the hypothesis is tested.
Tell a teammate the boundary out loud
If you cannot explain the boundary in one sentence to a person who does not care about agents, it is not a boundary. It never sends is a boundary. It is careful is not. The trial is short enough that sloppy language becomes a send. Write the sentence on the charter. Read it. Then start.
The same sentence should appear in the first message you send the bot, not only in a file you intend to paste later. Trial users under-specify because they are exploring. Exploring is how a browser session becomes a login. Exploring is how a "just checking Gmail" becomes a send. The metered credit makes that worse, because there is no second week to notice. There is only the packet, the preview, and whatever left the building.
If you are evaluating with a colleague, split the roles. One person writes the brief. The other person watches the Agent Computer and is allowed to abort. A trial run with nobody watching the desktop is not a test of the product. It is a test of how much you trust a system you have not measured. Trust is the thing the trial is supposed to earn. Do not donate it in advance.
When the run ends, compare the packet to the brief in the same sitting. If you wait until tomorrow, you will grade generosity, not compliance. Generosity is how "it was mostly right" becomes a production bot with the same sloppy charter. The trial's job is to make the charter stricter, or to tell you to walk away.
Copy the charter into a document you own before you close the app. If the trial expires and the bot disappears, you still have the brief. If you upgrade, you paste the same brief instead of inventing a sprawling one because you feel you should "use the plan." The document is the continuity. The product is the runtime. Mixing those is how trial slop becomes production slop.
One more pass: search your password manager for accounts created that afternoon. Trial users forget the dedicated inbox they opened just in case. That inbox is now a forwarding path onto a VM. Close it. Then search the VM's browser for saved passwords if the product UI exposes them. Assume they exist until you have evidence they do not. A trial is a lab. Labs get cleaned. A lab you do not clean is a second home for whoever still has the cookie.
A trial is a lab. Labs get cleaned. A lab you do not clean is a second home for whoever still has the cookie. If that sentence feels heavy for a sampler, you are ready to skip the product. The people who should skip it are not unserious. They are the people whose mail cannot live on a shared datacenter desktop for a week of curiosity. Choose the public-web brief, or choose not to trial. Those are the two adult options.
Keep reading: Is Grok Bot worth it is the verdict after a paid week, and supported platforms is the client list the trial will not expand.
Frequently Asked Questions
Is the Grok Bot free trial unlimited for seven days?
No. The 21 August 2026 trial is limited usage, consumed by work and tokens. Treat it as a small credit with a window, not as a full week of unrestricted bots. A single long browse can use most of it. Plan one reversible job, stop when you have a packet, and do not schedule a roster. Exact credit amounts are not on the SAFE list, so ignore blogs that print a number they cannot source.
Do I need a credit card to start the Grok Bot free trial?
Follow the current checkout in the official app. Some writeups say a card is collected up front. This page will not invent the billing-form details. What you can plan for is that there is no documented forever-free Grok Bot plan, and that after the metered trial the paid doors start at Cursor Pro+ at $60 a month for individuals. Read the screen in front of you, not a launch blog.
Can I try Grok Bot on Linux or Android during the trial?
Not with a native client. The FAQ lists macOS, Windows, and iPhone. It states there is no Linux desktop app. Android and iPad are unsupported. The cloud computer is a managed Linux VM, which is not a Linux desktop client. If you only have Linux or Android, the trial will not give you a supported way to operate the product.
What should I do after the Grok Bot free trial ends?
Either upgrade onto an eligible plan because one job earned it, or revoke every login you created and stop. Do not leave a Gmail session on a shared computer you no longer intend to pay for. Do not assume deleting the bot cleaned the VM. The cheapest documented individual paid path is Cursor Pro+ at $60 a month. Teams Standard is $40 a seat. Re-read vendor pricing the day you choose.