Know what you're paying for

Grok usage limits by plan

Paid tiers buy a larger allowance and higher priority rather than a stated number of messages. SuperGrok sits above X Premium, SuperGrok Heavy above SuperGrok, and the API is billed per token instead of per window, so the only figure that is official for you is the one in your own account.

There is no official message count for free, X Premium, X Premium+, SuperGrok Lite, SuperGrok or SuperGrok Heavy, and any page quoting one has either estimated it or is repeating a figure that has since moved.

What SpaceXAI documents

• API pricing and context length. Rates per million tokens, cached input pricing and the context window are published in the developer docs and change with a release note.
• Which plans exist. free, X Premium, X Premium+, SuperGrok Lite, SuperGrok, SuperGrok Heavy and the metered API.
• That consumer allowances are capped and refill on a rolling window rather than at a fixed hour of the day.
• That the same subscription behaves differently per surface: grok.com, Grok in X, the iOS and Android apps, Grok Imagine and Grok Build each have their own allowance.
• That the API is metered separately, at per-minute request and token rates, and is not subject to the consumer windows at all.

What SpaceXAI does not document

• Message counts for any consumer plan. No number is official for free, X Premium, X Premium+, SuperGrok Lite, SuperGrok or SuperGrok Heavy.
• Image and video generation allowances, which have moved repeatedly and were only ever visible through user reports.
• How load is prioritised between paid tiers when demand is high, which is why a working plan can still hit a wall.
• Advance notice. Quotas are adjusted silently, which is the reason this page is an event log rather than a table of current numbers.

Where to see your own limits

Inside Grok, on the account or usage screen for the surface you are on. It shows what is left on your own allowance and when it refills. That is the only authoritative figure for you, and it is the one number on this subject worth trusting.

Grok Imagine limits

Grok Imagine has its own allowance, counted separately from chat on every plan. SpaceXAI has never published a number for image or video generations; the allowances have moved repeatedly and were only ever visible through user reports, so each move is recorded in the dated log below with its evidence level.

The asymmetry that matters is per surface, not per plan. The same subscription can be generous on grok.com and tight in Grok Imagine, and an allowance spent in the app is not refunded on the web, which is why the history below records plan and surface separately.

Grok API rate limits and pricing

The API is metered separately from the consumer plans: per-token pricing and per-minute request and token rates, published in the developer documentation and changed with a release note. It is not subject to the consumer rolling windows at all.

The current rates sit with the release that introduced them — Grok 4.6 pricing is on the release page, dated and sourced. When the next release moves a rate, it lands there first and in the dated log below.

When did Grok's quotas last change?

See the full record in the limits changelog. We began monitoring SpaceXAI's public limits, pricing and plan pages on August 16, 2026; changes before that date are not in our archive.

When we knew what

A quota figure that we later corrected keeps its earlier reading here, with the evidence level and the source we held at the time. The log records what we believed and when, not only what stands today.

No corrections recorded since 2026-09-06. Every quota figure we have published still stands as first recorded.

What does each Grok plan include right now?

Every figure below is the most recent recorded change for that plan and surface, not a separate list we maintain by hand. If it is not in the change log, it is not on this page.

The full change log

The full change log, newest first. Most Grok quota changes are never documented by the vendor, so each row states how we know it — and a reported change is never dressed up as a documented one.

Documented

In the vendor's own documentation at the time.

Announced

Stated by the vendor in a post, not written into the docs.

Reported

Widely reported by users and not acknowledged by the vendor.

Inferred

Our own deduction from before-and-after observations.

• 2 November 2026API · Grok ImagineFeature availability — DeprecationDocumented

  available: grok-imagine-image-quality, grok-4.6, grok-imagine-image-2.0, grok-imagine-image, grok-imagine-image-pro→retired, served by grok-imagine-image-2.0

  xAI retires grok-imagine-image-quality from the API on this date; requests to the retired slug are served by grok-imagine-image-2.0 at low quality instead.

  In the vendor's own documentation at the time. Confidence 100/100. Source

• 12 August 2026API · APIContext window — AddedDocumented

  Not offered→500k tokens

  Grok 4.6 raised the context window to 500k tokens.

  In the vendor's own documentation at the time. Confidence 95/100. Source

• 12 August 2026API · APIPrice — Price changeDocumented

  Grok 4.5 pricing→$2 in / $0.50 cached / $6 out per 1M under 200k prompt; $4 / $1 / $12 above USD / 1M tokens

  Grok 4.6 launched with a two-tier price structure that switches at a 200k-token prompt.

  In the vendor's own documentation at the time. Confidence 90/100. Source

• 24 July 2026SuperGrok · iOS appFeature availability — DeprecationAnnounced

  3D companion mode available→Being retired

  The vendor announced the retirement of 3D companion mode; the rollout ran account by account with no stated completion date.

  Stated by the vendor in a post, not written into the docs. Confidence 85/100. Source

• 15 May 2026API · APIFeature availability — DeprecationDocumented

  available: grok-4.6, grok-4-1-fast-reasoning, grok-4-1-fast-non-reasoning, grok-4-fast-reasoning, grok-4-fast-non-reasoning, grok-4-0709, grok-code-fast-1, grok-3→retired, served by grok-4.3

  xAI retires May 15, 2026 Model from the API on this date; requests to the retired slugs are served by grok-4.3 instead.

  In the vendor's own documentation at the time. Confidence 100/100. Source

• March 2026SuperGrok · Grok ImagineVideo generation — DecreaseReported

  Dozens per day→Roughly 10 per 8 hours videos / 8h

  Users reported video quotas cut by up to 80 percent with longer reset timers; the vendor did not publish the change.

  Widely reported by users and not acknowledged by the vendor. Confidence 70/100. Source

• March 2026Free · Grok ImagineVideo generation — RemovedReported

  Daily free generations→None, subscription required

  Free video generation on Grok Imagine stopped; users reported a subscription became required.

  Widely reported by users and not acknowledged by the vendor. Confidence 70/100. Source

• Around March 2026X Premium · grok.comPriority access — RestrictedInferred

  Full access via X Premium→Deprioritised behind SuperGrok during peak load

  X Premium subscribers began hitting the high-demand wall on grok.com after SuperGrok launched as a separate tier.

  Our own deduction from before-and-after observations. Confidence 60/100. Source

Why does Grok say it is under high demand?

Most of the time that message is a plan limit being reached on your account, not a service failure. Quotas move without announcement, so we record each change as a dated event with an evidence level - documented, announced, reported or inferred - instead of asserting a single current number.

Before assuming an outage, check high demand against live status — the feed can report every surface as available while your own plan is throttled, and the outage history shows what an actual failure looks like.

Why estimates disagree

Every published estimate is derived, not disclosed. Two people on the same plan can get very different amounts of work from one allowance, because model choice, message length, tool calls, image or video generation and reasoning effort all draw on it differently. It is also why a change can be real and invisible at once: when demand is high, a plan that still works can be quietly deprioritised behind a higher tier.

Which Grok features have been retired?

Retired and removed features are recorded in the same quota log as quota changes, on the feature-availability dimension, dated and marked by evidence level. There is no separate graveyard page; the dated record is the answer.