Grok Bot's Quota Begins to Fall Off a Cliff; Musk Admits Subsidies Are Unsustainable

Over the past few days, Grok Bot users on X have noticed that quotas are suddenly being consumed much faster. Some woke up to find usage had jumped 20%, 30%, or even 40%. Quotas that used to last more than a week suddenly hit bottom. In the screenshot, user @DogecoinNorway asks Musk whether quotas can be reset, and Musk replies: the cost of providing Grok Bot is unsustainably high relative to revenue.

The reply is very direct. Translated into product language: the subsidy has stopped, or at least can no longer be provided in the same way.

A few days earlier, Grok Bot was powered by Opus 5.5, and a not-small share of tasks were routed to Opus 5.5. Later, users found that the chance of hitting Opus 5.5 was getting lower and lower, and the quality of the same task began to vary. The model is still the same entry point, but who is behind it may not be shown in the interface.

In the comments, someone compared this storyline with Cursor: first unlimited usage, then quietly changing routing, and finally deducting from a pool based on usage. Others mentioned that the worst part is not running out of quota, but having felt the day before that it was more than enough, only to open it the next day and see a chunk already gone. The letdown is even worse than actually running out.

These responses point to the same problem: the inference cost of a Personal Agent is not on the same order of magnitude as ordinary chat. The more active the user, the more background tasks, and the more the platform loses. Low-priced monthly subscriptions plus always-on calls to expensive models make the ledger hard to balance over the long term. Meta has ad revenue that can subsidize Muse, but as long as users keep it running in the background all day, costs will pile up the same way. The only difference is who can withstand it longer.

Another thing repeatedly mentioned is transparency. Some users want a receipt at the end of a task: which model was used and how much quota was deducted. Others want user-defined routing, or a more expensive explicit plan for power users. Still others directly say BYOK. People aren't unable to accept price increases; they just can't accept playing a guessing game with their balance.

Grok Bot's quota jump this time lays bare the business model of Personal Agents. Free quotas, low-priced subscriptions, always-on background operation, and routing to expensive models—put these together, and in the short term they can buy growth, but in the long term revenue has to cover inference costs. If it can't, the only options are to shrink quotas, downgrade routing, raise prices, or do all three at once.

The free lunch is still here, but it's starting to come with a time limit. When choosing this kind of tool, quota rules, model routing, and usage records are more worth looking at than a few highlight tasks in a demo video. After all, a demo only runs once; the ledger has to run every day.

Source: a tweet by Guizang (guizang.ai) https://x.com/op7418/status/2117092326690726075 , and Musk's reply https://x.com/elonmusk/status/2116992258520064397 .