xAI Quietly Drops SuperGrok Heavy's 'Near-Unlimited' Promise at $300

When xAI launched SuperGrok Heavy at $300 per month, the pitch was blunt and seductive: near-unlimited usage. That phrase did a lot of work. It was the single clearest justification for a price point that dwarfs most streaming bundles and rivals a car payment. As of this week, the phrase is gone from the pricing page. In its place: “highest usage limits.” The price did not move.

On the surface this reads like copy-editing. In practice it is a quiet renegotiation of what $300 a month buys — and it lands on top of months of shrinking quotas that Heavy subscribers have been documenting across forums and social platforms.

What actually changed

Two things, only one of which xAI announced in any formal sense.

First, the wording. The SuperGrok Heavy pricing page now promises the “highest usage limits” rather than “near-unlimited usage.” The distinction matters more than it sounds. “Near-unlimited” is a claim about freedom from limits. “Highest usage limits” is a comparative claim — you get more than the cheaper tier, but you are firmly inside a metered system. It is the difference between an all-you-can-eat restaurant and one that serves the largest plate on the menu.

Second, the real-world quotas. Community reports tracked through mid-2026 put the daily video generation allowance on Grok Imagine down from roughly 100 clips to around 30 — at the same $300 price. Subscribers describe burning through a fresh weekly quota within an hour or two of heavy Imagine use, then losing the ability to run ordinary chats without topping up.

The mechanism behind that squeeze is structural. Since June 2026, paid Grok plans run on one shared weekly usage pool spanning Chat, Imagine, Voice, and Build. The per-two-hour caps that most older guides still quote were retired at the same time. xAI’s own FAQ confirms the model: a single weekly budget, resetting on a schedule shown in the Usage tab, that every product draws from. A single video generation can consume a large slice of the same budget that your everyday questions depend on. Generate aggressively and you ration yourself out of chat.

xAI’s release notes from mid-August show the company at least acknowledging the experience problem — clearer notices when the pool runs dry in Imagine and Voice, plus a credit gauge in agent chat. Transparency about running out is welcome, but it is not the same as the original promise.

The tier almost nobody notices

Part of why this story matters: Grok’s consumer lineup has grown into six tiers across two parallel subscription paths — free, X Premium at $8, SuperGrok Lite at $10, SuperGrok at $30, X Premium+ at $40, SuperGrok Plus at $100, and Heavy at $300. Two mistakes show up repeatedly:

1. Paying $40 for X Premium+ without using X socially. It costs $10 more than SuperGrok while delivering less AI capability. Premium+ makes sense for people whose life is on X; for everyone else it is a trap.
2. Jumping from $30 straight to $300. SuperGrok Plus at $100 exists precisely for people who keep exhausting the cheaper plan, and it is the option most Heavy-curious users have never seriously evaluated. If your actual pain is running out of quota, tripling down to Heavy after a wording change deserves more scrutiny than it usually gets.

Six days to a double deadline

The SuperGrok story is also a useful reminder that this entire pricing layer is in motion industry-wide. Monday, August 31 brings two deadline events at once:

• Claude Sonnet 5 reprices — from $2 to $3 per million input tokens and $10 to $15 per million output, alongside a tokenizer change that adds 10–35% more tokens on code workloads. For teams running Sonnet 5 in production, the effective increase is larger than the headline rates suggest.
• GPT-5.4 and GPT-5.4 mini leave Codex for ChatGPT sign-in users (both remain available via API key), and Moonshot sunsets kimi-k2.5 and moonshot-v1 in favor of kimi-k3.

Further out: OpenAI’s Apple hearing lands October 1, DeepSeek’s current chat and reasoner models deprecate October 24, and GPT-5.6 promotional pricing ends around November 21. Anyone budgeting AI spend for Q4 is doing arithmetic on a shifting floor.

And on the Grok roadmap itself: Musk’s stated window for Grok 4.7 is early-to-mid September, with pre-training reportedly complete and supplemental training running on SpaceX engineering data — though no model ID has appeared at docs.x.ai yet. Heavy subscribers paying $300 in hopes of day-one access to the next flagship are betting on a roadmap promise from a company that just quietly edited an older one.

Why this pattern keeps repeating

The deeper story is that “unlimited” was never a sustainable promise for frontier AI subscriptions. Inference at frontier scale costs real money per token, and multimodal generation — especially video — costs orders of magnitude more than chat. Every lab that promised abundance has converged on metering: OpenAI, Google, and Anthropic all walk the same line between marketing generosity and unit economics. xAI is not an outlier in having limits; it is only an outlier in how prominently “near-unlimited” featured in its top-tier pitch, and how quietly that pitch changed.

For subscribers, the practical takeaways are unglamorous but real: check the Usage tab before assuming your plan behaves like it did at sign-up, model your actual Imagine and Voice consumption against the weekly pool, and give the $100 tier an honest look before assuming Heavy is the only escape hatch from rate limits.

A promise removed from a pricing page is a small thing. A subscription that costs $3,600 a year changing what it delivers while the price holds steady is not. Heavy subscribers just learned the difference.