Grok Bot Reaches Cheaper Grok and Cursor Plans, but the Limits Are Real

The most significant change here isn't the price list. It's that Grok Bot has moved down the ladder. Until recently, the agent sat mostly in the pricier tiers. Now it's available on the cheaper plans too: SuperGrok, Cursor Pro, and every Cursor Teams option, while still remaining in the top-end bundles. According to SpaceXAI, X Premium+ subscribers can also link their accounts to reach the tool. That opens several new routes in, and it's a much wider door than the one available on launch day.

The reason for the overlap is corporate, not product-driven. In April 2026, Cursor agreed a tie-up with SpaceX to speed up model training on SpaceXAI's Colossus infrastructure. A few months later, in August 2026, SpaceX moved to buy Cursor outright, and Grok Bot was formally introduced around the same time. Cursor had also just launched its Origin platform as a challenger to GitHub. If you're wondering why the same agent now shows up under Grok branding and Cursor branding, that history explains it.

What each tier costs

On the Cursor side, the Pro plan is $20 a month (about €18 at the current exchange rate, before taxes), Pro+ is $60 (about €54), and Ultra is $200 (about €179). For teams, Teams Standard costs $40 per user per month (about €36) and Teams Premium rises to $120 (about €107). Members of the direct-join team plans get automatic access to the system, so there's no per-person setup beyond joining.

On the Grok side, SuperGrok is $30 a month (about €27), SuperGrok Plus is $100 (about €90), and SuperGrok Heavy sits above those, though the source doesn't give its price. The alternative route is linking an X Premium+ account, which costs $40 a month in the United States (about €36). Enterprise customers on both Grok and Cursor Enterprise can use the agent through their organisation, or ask their sales contact to switch it on.

My read: for a solo developer, the decision comes down to whether you already pay for X Premium+ or a Cursor plan. If you do, you're probably one settings change away from trying the agent. If you don't, the $20 Cursor Pro tier is the lowest-friction entry, but the usage rules below matter more than the sticker price.

The weekly quota is the catch to plan around

A monthly fee doesn't buy unlimited use. The quota included in each plan resets every week. Anything beyond that only continues if you've turned on shared on-demand spending in your account. That's a real cost decision, not a footnote, especially for anyone planning long, unattended jobs.

There's also a seven-day free trial, but it's usage-based rather than capped by a set number of tasks. Cursor doesn't publish a fixed allowance for the trial. Instead, it meters credit by the steps the agent takes and the volume of tokens processed (the units of text and data the model reads and writes), rather than by counting messages. If you're used to thinking in chat turns, that's a different mental model. A single request that triggers a long browsing session can burn far more than a simple question would.

What the cloud agent actually does

This isn't a chat window with a better model behind it. The system is built to hand off work that runs on its own. Each Bot operates on a persistent computer hosted in Cursor's cloud. It has direct access to a web browser, a file system, and a terminal, which is the command-line interface for running code, managing files, and launching programs.

The workflow is straightforward in principle. You define the outcome you want, grant the tools it needs, and the task keeps running on remote servers even after you shut down your own machine. The agent can carry context from one conversation into another and repeat workflows on a schedule. SpaceXAI describes these Bots as AI teammates that complete tasks from start to finish, stopping only when a step needs a human to sign off.

The use cases the company points to are practical and fairly narrow: researching potential sales prospects, pulling recent activity from a CRM (customer relationship management software), and drafting personalised messages that wait for review before anything is sent. It can also reproduce software bugs, generate reports on customer status, or pull together executive summaries from email, calendars and other connected sources. Those are repetitive, multi-step chores, which is where an always-on agent should earn its keep.

Where it still falls short

The tool is still in beta, and the limits are not minor. Many websites block automated browsing, show CAPTCHA checks, or require logins and other steps that only a human can complete. That means a workflow that looks clean on paper can stall halfway through a real site. Higher-impact actions are also held in an approval queue before the Bot can proceed. That's a sensible safety gate, but it means the agent is not fully hands-off.

Early users also report that some workflows still need more hands-on supervision than the pitch of leaving the agent to work alone implies. If you're evaluating it, I'd start with a bounded, low-risk task you can check easily: a research pass, a bug reproduction, or a draft that you review before anything goes out. Those will show you quickly whether the agent saves time or simply moves the babysitting somewhere else.

What to watch next is whether the browsing restrictions ease and whether the approval queue becomes more selective as the product matures. Until then, the wider availability is good news for access, but the real question is how much work you can hand off without standing over it.