Elon Musk has said that the cost of serving Grok Bot was "unsustainably high relative to revenue." The remark, posted on X on November 1, 2026, was a reply to a user who asked whether the weekly usage pool could be reset after many people ran out quickly, according to Startup Fortune. It is the only primary statement so far. Musk gave no figures, no new limits and no timeline.
Plan and usage changes hit paying users directly, so this deserves a careful read: what Musk said, what users report, what xAI has and has not published, and what builders can do today. For the background, see our post on the Grok Bot early beta.
TL;DR: claimed versus verified
| Claim | Source | Status |
|---|---|---|
| Serving Grok Bot costs "unsustainably" more than it earns | Musk on X, Nov 1 | Said by Musk; no figures |
| Weekly limits were cut | Users on forums | Users report fast depletion; no official change found |
| Grok Bot has about 800,000 paid weekly users | Teslanorth via Startup Fortune | Unverified, secondary report |
| Grok Bot has no standalone price | Startup Fortune | Reported; check official plan pages |
| Weekly allowance size | Not published | Unknown |
| A fix or reset is coming | Not stated | Unknown |
What exactly did Musk say?
The quote, as Startup Fortune and other outlets relay it from his X post: "The cost of serving Grok Bot was unsustainably high relative to revenue." The post was a reply to a user asking whether xAI could reset the weekly pool after "lots of people" ran out unusually fast that week.
What it does not say matters as much. It gives no serving cost, no revenue, no old or new limit values, and no date for any change. It does not say Grok Bot is shutting down or that prices will rise. Treat it as an acknowledgment of cost pressure, not a policy announcement.
What are users reporting?
The clearest user data we found is in a Cursor community forum thread, where Grok Bot is also offered through Cursor plans. These are individual reports, not verified measurements.
- A user on the $200 Ultra plan reported the weekly allowance at 65% on day one, 84% on day two and 99% on day three. Their export for days two and three showed about 284 million tokens, roughly 84% of them cache reads.
- A user on the $300 SuperGrok Heavy plan reported using 70% of their Grok Bot quota after two and a half days while using only 6% of the SuperGrok Heavy quota.
- Related threads report that weekly usage can spill into paid On-Demand without warning, and that usage feels low even on Ultra.
These threads date from September and October, so the depletion complaints predate Musk's remark. That is consistent with a cost problem that has existed for weeks, though it does not prove a recent limit change.
Why does the meter drain so fast?
A forum reply from an account that appears to be staff explains the mechanics: Grok Bot bills for the bot's work, meaning tool calls, browser use, subagents, and re-reading conversation history on every turn, not for the number of messages you send. The same reply says idle old conversations do not drain the allowance, and cost applies only when a new message is sent in that chat.
That fits what users describe: the whole chat history is re-sent with each message, so a long thread gets more expensive every turn. A routine that runs every few minutes, the staff reply notes, fires about 100 times a day. Our Grok Bot guide for teams covers how routines and skills are set up, and our piece on the verification tax shows a related cost of agent work: checking and redoing output.
How can you stretch your allowance?
Practical steps reported in that thread:
- Start a fresh conversation per task. Long chats re-send history each turn.
- Space out routines. Fewer scheduled runs, longer intervals.
- Put recurring jobs on a dedicated bot so they do not share context with interactive work.
- Delegate heavy implementation to Cursor cloud agents, per the staff-style advice.
- Scope tasks tightly. Narrow instructions mean fewer tool calls and subagents.
- Watch for On-Demand spill-over. Users report usage can move to paid overage without a warning, so check billing settings.
Our posts on real-world Grok Bot use cases and a solo founder's week with Grok Bot show which workloads are heavy.
How should paying users read a cost admission?
A CEO saying a product loses money at its current price can mean several different things, and the post gives no way to tell them apart. It could precede tighter weekly pools, a new higher tier, usage-based billing, a cheaper default model, or nothing at all. Past examples across the industry show all of these outcomes, so the useful move is to prepare for the range rather than guess one.
Three questions help you decide how exposed you are. First, how much of your work depends on background routines, which are the heaviest consumers according to the forum guidance? Second, do you have a second tool that can take over if the pool runs out mid-week? Third, are you on a plan where overage bills automatically, which users say can happen without a warning?
If the answers are "a lot," "no" and "yes," treat Grok Bot as a metered utility and monitor it weekly. Export usage if the product lets you, as the forum user did, and compare token counts against the work actually delivered. A cache-read share near 84%, as one user reported, tells you most spend is repeated context, which you can reduce with shorter threads and tighter scoping, not by sending fewer messages.
Teams should also record the plan terms they bought under. If limits change materially, a dated record of the allowance, price and features at purchase time supports a request for a refund or a plan change, and it makes comparison with alternatives honest.
What do we know about pricing?
Startup Fortune reports that Grok Bot has no standalone price. It comes with SuperGrok, from $30 a month, or with Cursor plans, each carrying a weekly compute allowance that xAI has not published. It also reports no free tier for the agent. We could not verify these from an official xAI pricing page in this run, so confirm current terms on the plan pages before deciding.
Another outlet-level claim, about 800,000 paid weekly users, traces to Teslanorth via Startup Fortune and is not corroborated by a primary source. The same article says Musk earlier floated ads in Grok answers to help pay for "those expensive GPUs." Treat both as context, not confirmed figures.
What does this say about agent economics?
Agents are expensive to serve because each task is many model calls. Tool use, browsing and subagents multiply tokens, and cached reads still cost something. A flat subscription with a weekly pool meets that cost curve when heavy users run background routines all week. Several vendors have moved toward tighter metering, and Startup Fortune says OpenAI has begun metering token usage more tightly across paid tiers, a claim we did not verify.
The lesson for builders is to design for cost: compact context, scope tasks, and measure tokens per job. Our OpenAI Dots versus Grok Bot comparison covers how the personal-agent products differ in packaging.
Safety note for always-on agents
Cost is not the only risk of background agents. If you let an agent hold logins and run unattended, limit what it can do. See our post on Grok Bot's own email address and approvals. AgentBeam is the agent security platform from the explainx.ai team that stops AI agents before they take dangerous actions.
What to watch next
Because the facts here are thin, the next few days will decide whether this was a passing reply or the start of a plan change. Keep an eye on the places where a change would show first: xAI's own plan and docs pages, the Cursor pricing and forum pages where Grok Bot is bundled, and Musk's account for follow-up replies. Compare any new numbers with the user reports above before accepting them.
- An official limit statement on xAI's plan pages or Grok Bot docs.
- A reset or plan change, which Musk was asked about but did not promise.
- Published allowance sizes. Until xAI discloses them, users cannot plan.
- Pricing changes to SuperGrok tiers or Cursor bundles.
Bottom line
Verified: Musk said on X that serving Grok Bot costs too much relative to revenue, and users on several plans report exhausting their weekly allowance in days. Not verified: any specific limit cut, the size of the allowance, the user count, or a fix. If Grok Bot is part of your workflow, treat the allowance as a budget, track what drains it, and keep a fallback.
Related reading
- Grok Bot early beta: persistent AI agents
- Grok Bot guide: teams, skills and routines
- Grok Bot real-world use cases
- A solo founder's week with Grok Bot
- Grok Bot gets its own email address
- OpenAI Dots vs Grok Bot vs Meta Muse
Details are accurate as of November 2, 2026. Musk's post is relayed from Startup Fortune and we could not open it directly; check plan pages for current limits.