Grok's Weekly Usage Pool Explained: How One Video Can Use Up Your Weekly Chat Quota

You generated three videos on Tuesday morning. By Tuesday afternoon Grok told you that you'd hit your limit — not just for video, but for everything. Chat stopped. Image generation stopped. And the reset wasn't tomorrow; it was next week. If that's the situation you're trying to understand, this article explains exactly what happened.

The short version: in June 2026, xAI replaced its per-product daily rate limits with a single shared weekly usage pool. Before that change, video had its own allowance, chat had its own allowance, and running out of one didn't touch the other. After the change, there is one weekly allowance and every Grok surface draws from it.

That is a genuinely different product. It's also poorly explained by the interface, which is why so many users experience it as their quota mysteriously vanishing. Below is what changed, why video hits so much harder than chat, what xAI has and has not published, and the practical options if you need both a working chatbot and a working video generator in the same week.

What Actually Changed in June 2026

The old system was straightforward, if restrictive. Each product had its own meter and its own reset cadence — mostly daily. If you exhausted Grok Imagine's daily image allowance, you waited until tomorrow, and in the meantime Grok chat kept working normally. Limits were annoying but they were compartmentalised, and they were short.

The new system collapses all of that into one number:

Two things got worse at once, and it's worth separating them because people often blame the wrong one.

First, the pooling. Your most expensive activity now determines whether your cheapest activity still works. Video generation is enormously more compute-intensive than a chat turn, so a video session doesn't just reduce your remaining chat capacity — it can eliminate it.

Second, the reset cadence. Moving from daily to weekly turns a bad afternoon into a bad week. Under a daily meter, over-generating on Tuesday costs you Tuesday. Under a weekly meter, over-generating on Tuesday can cost you Wednesday through Sunday as well. The failure mode got roughly seven times more expensive in lost time.

Everything That Draws From the Same Pool

Under the current system, the shared weekly allowance covers Grok's surfaces collectively. Chat, image generation and video generation all deduct from it. What xAI has not published is the exchange rate — how much of the pool any single action consumes.

The “relative weight” column is an inference from how compute works, not a figure xAI publishes. Rendering several seconds of video requires orders of magnitude more computation than producing a paragraph of text, and pooled quota systems price roughly in line with compute. That inference is consistent with what users report — but it is an inference, and you should treat anyone quoting you an exact “one video = N chats” number with suspicion, because that number has never been published.

Why One Video Can Eat the Whole Week

Here's the mechanism, step by step, in the order it usually bites people:

• 1.
  Video is priced against chat, not against video. Because everything shares one pool, the unit you spend on a render is the same unit you spend on a question. There is no separate video budget to run down first — the render comes straight out of the general allowance.
• 2.
  Video generation is iterative by nature. Almost nobody gets the shot they wanted on the first prompt. Realistically you generate, look, adjust the prompt, generate again. Three or four attempts to land one usable clip is normal. Every attempt spends pool.
• 3.
  Failures reportedly still cost you. Users consistently report that generations blocked by content moderation, or that error out mid-render, still register against the allowance. You lose the quota and receive nothing. On an iterative workflow this compounds fast.
• 4.
  The meter gives you no forward visibility. You can see roughly how much you have left. You cannot see what the next render will cost before you commit to it. So there is no way to make an informed decision about whether you can afford one more attempt.
• 5.
  When it's gone, everything is gone. The pool doesn't degrade gracefully by dropping video first and preserving chat. It empties, and Grok stops being useful for the rest of the week — including for the everyday text queries that cost almost nothing.

What xAI Publishes — and What It Doesn't

This is the part that makes the system genuinely hard to plan around. xAI publishes the prices clearly. It does not publish the pool sizes for any tier.

Every row in the pool-size column says the same thing, and that is the finding. You can compare prices across seven tiers, but you cannot compare what those prices buy, because the quantity being sold has never been quantified. The only comparative claim available is ordinal: Heavy's pool is larger than Plus's, which is larger than SuperGrok's. By how much is unpublished.

This is also the context for SuperGrok Heavy's marketing shift from “near-unlimited” to “highest limits.” The earlier phrase promised an experience; the current one promises a ranking. We covered that change and the $300 tier in detail in SuperGrok Heavy at $300: What “Highest Limits” Actually Gets You.

And note what the top row means for anyone hoping to test video before paying: there is no free Grok Imagine tier. A free X account gets chat only. Video generation requires a paid plan starting at SuperGrok Lite at $10/mo. Free-tier users have their own separate set of complaints about the chat meter, which we cover in Grok's Free Tier Quota: Reset Delays and Usage Complaints.

How to Tell If the Pool Is What's Hitting You

Before changing anything, confirm the diagnosis. Pooled-quota exhaustion has a recognisable signature:

If you're in the left column, no amount of prompt tuning fixes it. The constraint is structural: you are buying video generation out of a general-purpose allowance, and video is the most expensive thing in the store.

What You Can Actually Do About It

Five options, roughly in order of how much they cost you. The honest answer is that the first four are mitigations, not fixes — the fifth is the only one that removes the constraint.

• 1.
  Front-load your week deliberately. Decide before you start how much of the week's pool is for video and stop at that line. This is rationing, and it works, but it means treating a subscription you already paid for as a scarce resource.
• 2.
  Prompt on paper first. Since iteration is what drains the pool, do the iteration somewhere free. Write and refine the prompt fully before the first render, so you spend pool on execution rather than exploration.
• 3.
  Keep a second chat tool. If the thing you can't afford to lose is everyday chat, don't let it live in the same pool as your video work. Any separate assistant insulates your text workflow from your creative one.
• 4.
  Upgrade a tier. This buys a larger pool of unpublished size for a published price. It changes when you hit the wall, not whether. At $300/mo for Heavy, it's an expensive way to defer the same problem.
• 5.
  Take video out of the pool entirely. Generate video somewhere that charges per generation instead of metering a shared allowance. Your Grok subscription then goes back to being purely a chat subscription — which is the thing it's reliably good at — and your video work stops being able to break it.

Option 5 in Practice: Move Video Off the Pool

AIVeed is a dedicated AI video generator with no subscription and no usage pool. You buy credits, a generation costs a fixed number of them, and nothing else competes for the balance — there is no chat product on the platform to drain it.

What a Generation Costs Without a Pool

Because credits map to generations at published rates, the per-video cost is simple arithmetic rather than an estimate:

Model choice is where the range comes from. The 120-credit standard rate covers the Veo3-class AIVeed model; Seedance V1 Pro at 480p/5s runs 30 credits for drafts and social clips, and images start at 10 credits (about $0.05). Available models include AIVeed (Veo3-based), Lip Sync, Grok Imagine, GPT Image 2, the Seedance family (2.0, 2.0 Fast, 2.0 Mini, 2.5, 1.5 Pro and V1 Pro) and Kling 3.0. Seedance 2.5 is live, with longer 20-second and 30-second generations. Full breakdown on the pricing page.

You Don't Have to Cancel Grok

This isn't an argument for dropping your subscription. Grok is a capable chat assistant and the pooled system only becomes a problem when video is competing with chat inside it. Remove the competition and the subscription gets better, not worse.

The split most people land on: Grok for chat, a dedicated tool for video. Your weekly pool then serves only the lightweight activity it comfortably supports, and you stop discovering on Thursday that Tuesday's renders ended your week. If you want the broader version of that argument, including the other rate-limit workarounds, see How to Escape Grok's Rate Limits.

Testing the split costs nothing. Signing up with Google gives you 200 free credits, enough for roughly one standard video or several cheaper Seedance drafts, with no credit card required. If the output works for you, packs start at $5 and payment goes through PayPal, PayU for INR, or Dodo.