Grok Bot: SuperGrok vs Heavy—any performance difference beyond usage?

Discussions
,
1

We’re comparing SuperGrok with SuperGrok Heavy specifically for Grok Bot—the autonomous agent with its own computer, routines and connectors—not Grok chat or its Heavy mode.

We contacted support and received a reply from Cursor’s AI Support Assistant stating that:

  • Both plans use the same first-party models, and Heavy does not add a more capable Bot model, extra reasoning compute or multi-agent mode.
  • There is “no documented lower latency, priority processing, or faster tool/computer execution” on Heavy while both accounts remain within their allowances.
  • Heavy’s documented advantage is a larger linked usage grant.

The reply linked to:

That is helpful. We’re looking for clarification on actual internal behavior and firsthand experience, rather than another comparison of published usage allowances.

For staff with product/engineering visibility: Can you confirm whether subscription tier changes the Bot’s model routing, reasoning budget, queue priority or execution resources while both accounts are within their allowances? In particular, does “no documented” speed advantage mean there is no plan-dependent priority or speed advantage internally? We do not need confidential implementation details.

For users who have personally used both plans: Did upgrading or downgrading produce a noticeable difference in either:

  1. Task quality: Better planning, fewer mistakes/retries, or successful completion of tasks the other tier struggled with?
  2. Speed: Shorter waits to start, faster responses/tool actions, or lower total completion time?

Reports of no noticeable difference are equally useful. Please mention the plans, approximate testing date and an example task. Informal observations are welcome; comparable timings or redacted screenshots would strengthen them. Please distinguish what you observed from assumptions about why it happened.

Thank you.

3

Hey, thanks for the detailed request.

On the staff side, the answer you got earlier is still correct. The documented difference between SuperGrok and SuperGrok Heavy for Grok Bot is the size of the linked usage grant. We don’t publish or guarantee any plan-based advantage in latency, queue priority, or the speed of tool or computer actions, as long as both accounts are within their allowances.

About “does no documented mean no internal advantage”, we don’t share internal implementation details like how models are routed or how resources are allocated. This also isn’t something we want to lock in, since the configuration changes over time and any specific claim about internal behavior would quickly go out of date. What you should rely on is the documented behavior. It’s the same first-party models, and the difference is the usage amount.

One clarification so your plan decision is based on the current setup. Right now, linking SuperGrok on any tier, including Heavy, is a usage grant for Grok Bot and does not change your Cursor plan. If you saw older materials about “free Cursor Ultra with Heavy”, that changed in late August and is no longer current. The current terms are always here:

From firsthand experience, this is a great question for the community. If anyone tested both plans and noticed a difference, or didn’t notice one, please share here with the plan names, the rough test date, and an example task. “No difference” reports are useful too.