FULL STORY

Grok Usage Quota Backlash: From Complaints to Musk's Admission

Users first suspected Grok had been quietly downgraded, then erupted over runaway weekly usage quotas. Musk admitted costs were unsustainable, but paid users kept hitting limits, fueling ongoing backlash.

2026-10-29 ~ 2026-11-08 · 8 episodes · 50 posts

Episode 1 · Users Suspect Grok Bot on X Was Quietly Downgraded (2026-10-29, 2 posts)

Multiple users on X report that the official Grok bot has suddenly become noticeably dumber, with weaker reasoning and instruction-following, suspecting xAI quietly swapped it to a lesser model.

Episode 2 · Grok Users Report Runaway Weekly Usage Quotas, Sparking Cancellations (2026-11-01, 10 posts)

From November 1–2, X users reported a surge of abnormal Grok weekly usage behavior: quota bars spiking to 100% with little or no activity, quotas exhausted days before reset, and chaotic reset behavior. Reports span all paid tiers, from Premium to the $300/month SuperGrok Heavy and $2,000/year plans; at least one Heavy subscriber cancelled. xAI has not publicly addressed the issue.

Confirmed

  • @RachelVT42 compiled multiple reports: one user went from 53% at bedtime to 100% on waking, with quotas consumed after only a couple of simple chats; xAI's chief-of-staff agent confirmed the usage cap was hit but offered no explanation.
  • @heyshrutimishra saw her weekly usage jump from 58% to 100% with almost no activity; another user reportedly went from 34% to 73% overnight.
  • @DogecoinNorway (44k followers) publicly asked Musk to reset quotas, hitting 100% four days before reset; many replies reported the same.
  • Greg Gaskell (shared by @sachinmaya1980) cancelled his $300/month SuperGrok Heavy subscription after usage jumped from 50% to 100% overnight with unchanged routines, followed by account lockout and unresponsive support.
  • @AlGrigor, on a $2,000/year plan, reported percentages still climbing rapidly in real time, implying an earlier promised fix had not landed.
  • @NickPassig reported confused resets (told "resets in two days" but it reset immediately; another account burned 96% in a day) and said he would switch back to ChatGPT.
  • @nptacek noted rate limits keep changing weekly, especially after modes merged into one usage pool; a few images and short videos consumed 10% of his weekly quota.
  • @nimaowji, out of credits with 3 days left before reset, called for raising Premium usage pools by 1.5x.
  • @djcows joked that the Grok bot drinks usage like freshwater, showing the issue has become a community meme.

Unconfirmed

  • The root cause is unclear: a billing/statistics bug, hidden per-call consumption, or a silent quota cut. @AlGrigor's claim of a previously promised official fix rests on his own account alone.

Why it matters

  • Grok is competing with ChatGPT for heavy users; quota stability and transparency directly affect retention. The issue reaches the highest-value tiers and comes with lockouts and silent support, pushing some users back to ChatGPT.
  • If quotas were silently cut, it raises trust concerns over paid subscription benefits; even as a bug, the volume of concurrent reports indicates broad impact warranting an official explanation.

Episode 3 · Musk admits Grok Bot costs unsustainable as user quotas slashed overnight (2026-11-02, 21 posts)

Musk publicly admitted on X that Grok Bot's serving costs are unsustainable relative to its revenue, adding that "saving money is pointless" and that he was merely "acting rationally." Per Forbes, this came on a Sunday, less than two weeks after he touted Grok Bot as the fastest-growing product he had ever seen. The statement matched users' collective experience: quotas shrank overnight, with usage meters jumping to 100% without warning. Bezos joined the exchange, suggesting a $500 premium tier with token-based overage billing rather than a simple price hike, sparking wide debate over the sustainability of subsidized AI subscriptions.

Confirmed

  • Musk confirmed in X replies that Grok Bot's costs are unsustainable relative to revenue, saying "saving money is pointless" and that he was "acting rationally" (m3, m9, m14, m16, m17; m11, m12, m13 cite Forbes, with m13 noting it happened Sunday).
  • Multiple users reported overnight quota cuts: meters jumping to 100% without warning, with equivalent usage consuming 20–40% more quota (m3, m4, m13, m16, m19, m20).
  • op7418 observed that after Grok Bot announced Opus 5.5 as its driver, many tasks did route to Claude Opus 5.5 early on, but the routing probability has steadily declined since (m4; m18, m20 reshare the same observation).
  • Bezos suggested a higher subscription tier ($500) with token-based billing beyond the quota (m6).
  • Replit CEO Amjad Masad reshared Musk's explanation and warned subsidized AI subscribers to "prepare to be rug pulled" (m1, m2, m9, m10; m10 amplified via GaryMarcus).
  • Adam McBride responded that despite upgrading his plan, Grok Bot still delivers notably less reasoning than competitors, while saying he hopes Grok wins (m7).
  • Developer arthurcolle accused xAI of unilaterally changing quota allocation after users signed up, calling it "not sporting" (m16).
  • RachelVT42 observed polarized quota experiences—some never hit the cap, others exhaust it in a day or two—and suggested merging a low-cost tier into X subscriptions with a lower weekly token cap for light users (m15).
  • Open-source community LottoLabs commented that Qwen 4 plus Hermes or omp could achieve the same results "without limits" (m5); PaulGug cited data privacy as his primary concern over cloud-hosted assistants, preferring Hermes as a frontend that taps the strongest model on demand (m8).
  • User Cortex's reply "It's fine bro, you won't need money in the future anyway" received 1.2K likes and was widely shared (m3).
  • French AI industry newsletter (emmanuelvivier) framed the episode as compute costs squeezing pricing and quotas across AI products (m11).

Unconfirmed

  • Rob LeCleRc's claim that AI's biggest zero-to-one growth window has passed is personal opinion (m20).
  • The assertion that the ad-supported free internet era is ending in favor of pay-as-you-go AI (m19), and the bubble scenario of VC money running out followed by price hikes, churn, and valuation crashes (m9), are commentator speculation without official confirmation.

Why it matters

  • Musk, as a first party, rarely admits cost-revenue imbalance in an AI product—and did so while tightening quotas less than two weeks after hyping record growth, offering direct evidence for the debate on burning-money AI business models.
  • Users' perceived quota cuts, suspected model routing downgrades, and Bezos's "premium tier + token overage" suggestion all point toward a shift from subsidized/free AI pricing to usage-based pricing, with tiered plans for heavy and light users likely ahead.

1 more related posts →

Episode 4 · Grok Usage Limits Seemingly Doubled, Then Reverted (2026-11-02, 3 posts)

Users spotted Grok weekly usage limits apparently doubling (93% dropping to 47%), then seemingly being reverted after backlash; xAI has not officially confirmed either change.

Episode 5 · xAI Tests Trading User Data for Extra Grok Usage (2026-11-02, 2 posts)

Unverified leaks and app preview screenshots suggest xAI is testing a mechanism where users who hit their 100% usage cap can share their interaction data for model training in exchange for extra free Grok usage.

Episode 6 · Grok Users Hit Weekly Usage Caps Despite $1,200/Year Top Tier Plan (2026-11-03, 2 posts)

A user reports hitting 100% of weekly usage limits on Grok's $1,200/year top-tier plan and is weighing an upgrade to the $2,000 tier, which he says is worth it—highlighting strong demand for xAI's higher-capacity plans.

Episode 7 · Grok Bot Weekly Quota Slashed, Sparking Paid User Backlash and Refund Demands (2026-11-06, 8 posts)

Starting November 6, large numbers of users complained on X that Grok Bot weekly quotas had been slashed drastically, with paying users hit hardest. Koltregaskes cited cases of users burning through 90% of their quota on day one, forced to cut numerous scheduled tasks and pause all project work just to survive the seven-day cycle; he lamented that Grok has become a "calendar bot" and that the £90 subscription is hard to justify. Top-tier subscriber Al Grigor reported having only 22% of quota left with 6 days remaining in the cycle, unable to cancel due to an annual subscription, with customer support refusing a refund.

Confirmed

  • Multiple users (nimaowji, koltregaskes, etc.) report that the Grok Bot usage pool in X Premium plans is severely insufficient: even after disabling other services and concentrating quota on Grok Bot, it can't last a week—nimaowji called it "more like a demo"
  • A SuperGrok Heavy ($300/month) user said usage jumped from 50% to 100% overnight and got locked out, with no human response from support; another user said a single team message consumed 4% of quota
  • Ethanniser reported: subscribed to the $30/month X Premium (page states it includes Grok Bot with 4x usage), but was asked to upgrade to Cursor Pro or link SuperGrok when using it
  • Multiple users (eterinquirer, developer mikepat711, etc.) observed that Grok has become noticeably "dumber" this week, making home automation workflows built on Grok Bot unreliable

Unconfirmed

  • The exact rules and scope of the quota cuts—xAI has given no official full explanation (m1 mentioned a response from Elon Musk, but its contents were not elaborated in the source material)
  • Whether Grok's intelligence has actually declined is currently based only on first-hand user observations

Why It Matters

  • The incident has triggered collective cancellation and refund demands from $300/month tier users, with annual subscribers being refused refunds, raising subscription-benefit fulfillment issues
  • Mikepat711's experience highlights the risk of betting personal infrastructure on a single AI vendor: providers can adjust quotas and model behavior without warning or explanation—a widespread caution for developers relying on AI for automation

Episode 8 · Grok 4.7 bottoms out in benchmarks as pricing draws criticism (2026-11-08, 2 posts)

Grok 4.7 scored just 35 points, ranking last on Artificial Analysis, prompting criticism from users that high pricing and tightened limits are undermining an otherwise well-engineered product compared to alternatives like Hermes.