# Grok limitations: weekly usage, the free version and risks for business use

> Grok limitations: a single weekly pool since June 2026, what happens when it runs out, free version limits, video, files and the risks for a business.

Source: Giallo Studio (giallostudio.it) · Page: https://giallostudio.it/en/blog/grok-limits-risks · Published: 2026-11-08 · Updated: 2026-11-08

**Grok no longer has daily limits on paid plans: since June 2026, chat, images, voice, Build and API draw on a single weekly usage pool, and when it runs out you buy extra credit, turn on automatic top-up or move up a plan; on the free plan, limits on Chat and Voice remain.** Beyond the technical limits, the Grok limitations a business needs to know include operational risks: inaccurate answers, knowledge frozen at May 2026 without active search, an agent (Grok Bot) where you can't choose the model. The information comes from the [Grok FAQ](https://docs.x.ai/grok/faq), the [plans documentation](https://docs.x.ai/x-subscription/pricing) and the [legal FAQ](https://x.ai/legal/faq), read on November 8, 2026.

This guide has two parts. The first explains usage limits, because they are the ones you hit first. The second covers risks, which you hit later and which cost more.

<div class="key-points">
<p class="kp-title">In summary</p>
<ul>
<li>A single weekly pool for all products on paid plans, replacing the old daily limits.</li>
<li>Included usage: Premium 4 times Plus, Super 15 times, Ultra 50 times.</li>
<li>Extra credit from $5, web only, valid for one year and pricier than included usage.</li>
<li>720p video up to the plan's cap, then 480p; uploads of up to about 100 files per message on the web.</li>
<li>Main risks: hallucinations, no model selection in Grok Bot, a cloud environment shared between Bots, data region not guaranteed via the API.</li>
</ul>
</div>

## How do Grok's usage limits work today?

The biggest change of 2026 is precisely how usage is counted. Since June 2026, according to the [FAQ](https://docs.x.ai/grok/faq), paid plans use a single weekly pool shared across Chat, Imagine, Voice, Build and API, instead of daily limits per product. The pool is also shared with Cursor, because The X Subscription brings Grok, Cursor and X together under the same subscription ([overview](https://docs.x.ai/x-subscription/overview)).

The documentation doesn't express the pool in messages or tokens, but in multiples of the Plus plan. Here is the scale as of November 2026, with prices from the [official page](https://x.ai/pricing), excluding VAT.

| Plan | Monthly price | Annual price | Included usage vs. Plus |
| --- | --- | --- | --- |
| Free | 0 $ | — | Separate limits on Chat and Voice |
| Plus | 8 $ | 80 $ | Base |
| Premium | 30 $ | 300 $ | 4 times |
| Super | 100 $ | 1.000 $ | 15 times |
| Ultra | 200 $ | 2.000 $ | 50 times |

Source: [docs.x.ai/x-subscription/pricing](https://docs.x.ai/x-subscription/pricing). The practical consequence of a single pool is that activities compete with each other: an afternoon spent generating videos reduces what's left for chat or for work in Cursor over the same period. Plan details are in our guide to [Grok pricing and plans](https://giallostudio.it/en/blog/grok-pricing-plans).

## What happens when you run out of usage?

There are three options, all described in the [FAQ](https://docs.x.ai/grok/faq):

<ol class="steps">
<li><strong>Extra credit.</strong> From $5 up, purchasable only on the web, expiring after one year.</li>
<li><strong>Auto Top Up.</strong> An automatic top-up that prevents interruptions, but needs to be kept under control.</li>
<li><strong>Change of plan.</strong> Moving to a higher plan increases the weekly pool.</li>
</ol>

The point not to overlook is price: extra credit is charged at the standard rate, so it costs more than the usage included in the plan. Auto top-up turned on across several company accounts is the classic cost nobody sees until the bank statement arrives. It's one of the cases we describe in [hidden costs of AI projects](https://giallostudio.it/en/blog/hidden-costs-ai-projects).

Another rule worth knowing: anyone on a legacy plan (SuperGrok, SuperGrok Plus, SuperGrok Heavy or X Premium) keeps it until they change it, but switching to The X Subscription is permanent and can't be undone. And API credits, once purchased, are non-refundable.

## What are the limits of Grok's free version?

Grok “is free to get started,” according to the [official overview](https://docs.x.ai/grok/overview), and the FAQ specifies that limits on Chat and Voice remain on the free plan. The documentation read on November 8, 2026 gives no precise numbers. Many sites report a number of messages every few hours: without an up-to-date official source, we don't repeat it.

For business use, there's a limit that matters more than the number of messages: on the free plan, as on the other consumer plans, conversations can be used to train models until you turn off the “Improve the model” option ([legal FAQ](https://x.ai/legal/faq)). We cover this topic in depth in [Grok and GDPR](https://giallostudio.it/en/blog/grok-privacy-gdpr).

## Are there limits on images, video and files?

Yes, and they differ by content type.

- **Video.** Each plan has a cap on 720p videos; once you reach it, videos drop to 480p ([FAQ](https://docs.x.ai/grok/faq)). The pricing page lists 15-second 1080p videos for the Super plan.
- **Images.** They consume the same weekly pool as everything else. The “grok” watermark can't be removed.
- **Files.** Up to about 100 files per message on the web and 20 on Android; most files up to 150 MB, in formats such as PDF, DOCX, XLSX, PPTX, CSV, images, audio, MP4 and MOV.

For anyone working on long documents, the file limit is rarely the problem. Checking what the model extracts from them is a bigger one.

## What limits does Grok Bot have?

Grok Bot, the agent that works on a cloud computer, has access limits even before usage limits. It's included in the Premium, Super and Ultra plans; Plus and Free don't include it ([Grok Bot FAQ](https://docs.x.ai/grok-bot/faq)). It's used with a Cursor account, from desktop apps for macOS, Windows and Linux and from companion apps for iOS 18 or later and Android 9 or later ([getting started guide](https://docs.x.ai/grok-bot/get-started)).

There's also a geographic limit that directly affects Italy: the @bot tag on X, which passes a post as a task to your own Grok Bot, is not yet available for X accounts based in the European Union, the United Kingdom, Australia, Iceland, Liechtenstein and Norway. What counts is the country registered by X, not your location or a VPN ([documentation](https://docs.x.ai/grok-bot/tag-on-x)).

Finally, logins: for SSO, passkeys and CAPTCHAs, the Bot hands control back to the user. So a workflow that depends on these steps won't run on its own overnight. The full guide is in [Grok Bot: what it is](https://giallostudio.it/en/blog/what-is-grok-bot).

## What doesn't Grok know?

A model knows what it saw during training. For Grok 4.7, the flagship model, the knowledge cutoff is May 2026 ([model card](https://docs.x.ai/developers/grok-4-7)). In the app the problem is smaller because Grok can search the web and public X posts in real time. Via the API, however, the [models documentation](https://docs.x.ai/developers/models) is clear: without the Web Search or X Search tools turned on, Grok doesn't know about events after its training data.

Anyone building an integration who forgets to turn on search gets confident answers about outdated facts. What the cutoff means and how to handle it is explained in [knowledge cutoff: what it means](https://giallostudio.it/en/blog/what-is-knowledge-cutoff).

## What risks does a business using Grok face?

The following risks aren't our own assumptions: they come from the vendor's documentation. For each one, we show how we handle it in our projects.

| Risk | What the documentation says | How to reduce it |
| --- | --- | --- |
| Inaccurate answers | The legal FAQ warns that Grok may give inaccurate information or “hallucinations” and asks users to verify it | Mandatory sources and human review of numbers, regulations and names |
| No choice of model | In Grok Bot the system uses a SpaceXSI or sub-processor model for each task, without showing which | Keep on Grok Bot only tasks where variability is acceptable |
| Shared environment | A user's Bots share a persistent cloud computer; the docs warn not to use them as a security boundary | One account per scope, never production credentials in the same environment |
| Expiring approvals | Approval requests triggered by background work expire after about 30 minutes | “Ask first” rules on sending, purchases, deletions and permissions; someone who monitors them |
| Data region | The main API endpoint doesn't guarantee the region; no EU endpoint is listed | No personal data via the API without a transfer assessment |
| Public links | Shared links can be read by anyone who has them and indexed if published | Ban on publishing links to conversations |

Sources: [legal FAQ](https://x.ai/legal/faq), [Grok Bot FAQ](https://docs.x.ai/grok-bot/faq), [approvals and security](https://docs.x.ai/grok-bot/approvals-security-and-privacy), [regional endpoints](https://docs.x.ai/developers/advanced-api-usage/regions).

The risk of hallucinations isn't specific to Grok, but a vendor that states it in its own FAQ is a reminder that verification is the job of whoever uses the tool. The techniques we use to reduce it are in [AI hallucinations: how to avoid them](https://giallostudio.it/en/blog/ai-hallucinations).

## How do you decide whether Grok is right for a process?

With the same method used for any AI tool: start from the process, not from the tool. Before adopting it, we check these points:

<ul class="checklist">
<li>How many users, on which plan, and how much weekly pool is really needed.</li>
<li>Who can activate extra credit and automatic top-ups.</li>
<li>What data goes into conversations, and with which training setting.</li>
<li>Whether the task needs up-to-date data, and therefore active search.</li>
<li>Which Grok Bot actions always require human approval.</li>
<li>What happens if the vendor changes plans, names or models, as happened several times in 2026.</li>
</ul>

The complete method is in [how to assess AI risks in your company](https://giallostudio.it/en/blog/ai-risk-assessment); to start from the basics, see our guide to [what Grok is](https://giallostudio.it/en/blog/what-is-grok-ai).

## Limits and costs set by the company, not by the plan

A subscription with a weekly pool works for one person. When AI works inside a process, the company needs to decide for itself the spending cap, the model, where the data goes and which steps require a person.

That's why, when we build an agent at Giallo Studio, we don't accept limits imposed by a plan: we write them together with the client, as a cost cap per task, approval rules and an action log, on the model chosen for that job. If you want to understand which processes can move off a general-purpose subscription, let's start with [AI consulting](https://giallostudio.it/en/ai-consulting); if your goal is already clear, see how we work on [AI agents](https://giallostudio.it/en/ai-agents) or [write to us](https://giallostudio.it/en/contact).

## Frequently asked questions

### Does Grok still have daily limits?

Not on paid plans: since June 2026, daily per-product limits have been replaced by a single weekly usage pool shared across Chat, Imagine, Voice, Build and API. On the free plan, limits on Chat and Voice remain.

### What happens when you hit Grok's limit?

You can buy extra credit, starting at $5, only on the web and expiring after one year; you can turn on automatic top-up (Auto Top Up) or move to a higher plan. Extra credit is charged at the standard rate, so it costs more than the usage included in the plan.

### How much can you use Grok for free?

Grok is free to get started, with limits on Chat and Voice. The official documentation read on November 8, 2026 doesn't publish precise numbers for the free plan, so be wary of anyone quoting a fixed number of messages without citing a source.

### How much can you use Grok on paid plans?

The documentation expresses included usage relative to the Plus plan: Plus is the baseline, Premium is worth 4 times as much, Super 15 times and Ultra 50 times. As of November 2026, monthly prices are $8, $30, $100 and $200, excluding VAT.

### Does Grok know today's news?

In the app, it can search the web and public X posts in real time. Via the API, however, without the Web Search or X Search tools turned on, Grok doesn't know about events after its training data: for Grok 4.7 the cutoff is May 2026.

### What are the risks of using Grok in a business?

Answers can be inaccurate, as the official FAQ itself warns; in Grok Bot you can't choose the model and a user's Bots share a cloud computer; background approvals expire after about 30 minutes; via the API the processing region isn't guaranteed. These risks are manageable if you know them in advance.
