Seeing whether it can navigate your repo
Free
You have a real project and a terminal, and you want to know whether the thing is useful before it is attached to a card.
$0/mo
Costs nothing, and unusually for this section that is not a trial with a clock on it. The free plan is a standing tier that carries Grok Build, so there is no day on which access lapses. What there is instead is an undisclosed ceiling: the matrix says “Limited” and nothing more, the faster Grok 4.7 Fast serving is excluded, so the only way to learn how far it goes is to run it until it stops.
The agent on the smallest bill
Plus
You want Grok Build fully included and have no use for the bots, the video generation or the third-party models in Cursor.
$8/mo
$7/mo billed annually
The cheapest paid tier, and the unit every other tier is measured in: Premium is 4x this pool, Super 15x, Ultra 50x. Worth understanding what you are buying: not a Grok Build allowance but a general one, shared with Chat, Voice, Imagine and Cursor's agent when it runs Grok models. Plus also carries X benefits you may not want — a checkmark, creator tools, fewer ads — because the plan is sold as one subscription.
Long sessions and parallel subagents
Super
You run plans across large trees and let subagents work in their own worktrees, so several contexts are alive at once.
$100/mo
$84/mo billed annually
The tier the vendor recommends for daily agent users: 15x the Plus pool and priority model access at peak times. The step from Premium is 4x to 15x of a quantity never published, so the jump is knowable as a ratio and not as an amount of work. If Grok Bot is also running on the same plan, remember that its routines draw on this pool too.
The ceiling, for people who keep hitting one
Ultra
You exhausted the pool on the tier below more than once and would rather not think about it again.
$200/mo
$167/mo billed annually
The top individual tier, 50x the Plus pool, with earliest access to new products and models. It costs less than the top of the old lineup, SuperGrok Heavy, did when we checked in September, and its price sits on the pricing page in plain text — the old ceiling's was printed only inside an animated counter on another page.
A team with code it cannot leak
Teams
Several engineers, a private codebase, and somebody who has to answer for where the source goes.
$30/seat/mo
The reason to be on this row is not the seat management — it is the data controls. Teams adds team-wide privacy settings, and the Build documentation says zero data retention is enforced at team level once enabled. The consumer FAQ's line that content is not used for training is written about business and enterprise customers; the vendor does not say in so many words that a Teams seat on The X Subscription is one. For a tool pointed at a private repository, ask before assuming.
CI, scripts and anything unattended
xAI API key
Grok Build runs headless in a pipeline, reviewing diffs or triaging failures, with no one watching the terminal.
At model rates
A key detaches you from the pool entirely — the pricing page says API usage is billed separately and does not draw from it — and is the only route with a published rate. It also exposes the rule that matters most to an agent: crossing 200k prompt tokens re-prices the whole request at the higher tier, not just the tokens past the line. An agent accumulating repository context walks toward that threshold all session, and steps over it once — after which every token in that request, including the first, bills at the higher rate.