Grok Build: Free Access, App Creation, Publishing, Models, and Current Limits
- 56 minutes ago
- 6 min read
Grok Build started as a terminal tool for professional developers in May 2026. Three months later, it's a one-prompt app builder open to every Grok user, including free accounts.
........
Build Mode — the consumer, single-prompt app-creation layer — launched July 28, 2026 for SuperGrok Heavy subscribers only, then expanded to every plan, including Free, by August 25, 2026.
A prompt can produce landing pages, portfolios, business sites, planners and calculators, arcade and 3D games, and dashboards that pull in live external data.
Finished apps publish to a grok.me subdomain or a custom domain the user already owns, and the underlying code can be exported straight to GitHub.
The coding engine, grok-build-0.1, was open-sourced under Apache 2.0 on July 15, 2026 — two weeks before the consumer feature it powers even shipped.
xAI doesn't publish one fixed usage quota for Build; paid tiers draw from a shared weekly pool across Chat, Imagine, Voice, and Build instead of a separate daily cap.
··········
HOW GROK BUILD GOT HERE.
Build Mode is the consumer front end of a project that spent its first two months as a developer-only tool.
The original Grok Build launched in early beta on May 14, 2026 as a terminal-native coding agent aimed at professional software engineers — no chat window, no consumer framing.
On June 22, 2026, xAI added a /goal mode for autonomous multi-step execution, extending it from single edits toward longer implementation runs.
On July 15, 2026, xAI open-sourced the project entirely, publishing it to xai-org/grok-build under an Apache 2.0 license — a response, per the company, to user concerns about code syncing and privacy.
Build Mode, the one-prompt app-creation experience inside the consumer Grok apps, followed two weeks later on July 28, 2026. Read in order, the sequence is developer engine first, hardened in the open, consumer wrapper last — the reverse of how most AI app builders typically ship.
··········
WHAT A PROMPT CAN ACTUALLY PRODUCE.
xAI's own launch material lists specific output categories rather than a vague "build anything" pitch.
Supported categories include landing pages, portfolios, and business sites; productivity tools such as planners and calculators; arcade and 3D games; and interactive dashboards that can pull in external data.
At launch, xAI's team demonstrated the range with a driving simulator, a beat machine, and a physics playground — visual, interactive builds rather than another to-do list clone.
Under the hood, the agent can call on browser tooling to fetch live information and the Imagine image generator for on-the-fly visual assets within the same build session, rather than treating those as separate products.
··········
THE WORKFLOW: PLAN, REVIEW, APPROVE.
Grok Build doesn't silently rewrite a project — every change goes through a visible review step before it lands.
The agent works in a plan-review-approve cycle: it surveys the request, proposes a structured plan, accepts edits or a full rewrite of that plan, then executes and shows the result as a clean diff.
For larger jobs, it can fan out up to eight sub-agents in parallel, each working inside its own isolated Git worktree — closer to how a professional engineering team splits work than how a typical one-shot AI builder behaves.
For heavier reasoning steps beyond what the coding-specific model handles well, the agent can escalate to a larger general-purpose Grok model mid-session.
··········
PUBLISHING AND WHAT HAPPENS TO THE CODE.
This is where Build Mode breaks from being a closed sandbox demo.
A finished app can be published to a grok.me subdomain immediately, or connected to a custom domain the user already owns.
The full source code can be exported to GitHub, which turns Build Mode into a starting point for further development rather than a walled garden — anyone who outgrows the in-app editor can keep working on the exported repository elsewhere.
What's still unsettled, as of this writing: xAI hasn't published detailed hosting-uptime commitments, or clarified what happens to a live grok.me app if the underlying subscription lapses. Anyone publishing something meant to stay online should treat that as an open question rather than an assumption.
··········
THE MODEL BEHIND IT — AND A PRICING WRINKLE.
The engine's identity and its pricing both shifted between Build Mode's launch and the time of writing, and the two figures in circulation don't agree.
At Build Mode's July 28 launch, the coding engine was grok-build-0.1 (also referenced as grok-code-fast-1), carrying a 256,000-token context window, with API pricing reported at $1.00 per million input tokens and $2.00 per million output tokens.
A more recent source, checked in late August, describes Grok Build as now powered by Grok 4.6 — xAI's general flagship, released August 12, 2026 — at that model's own rate of $2.00 input, $0.50 cached input, and $6.00 output per million tokens, with a 500,000-token context window.
Those are two different price points for what's described as the same product, and the sources don't reconcile them. The most likely explanation is a genuine model swap sometime after launch, but that's an inference, not a confirmed fact — anyone budgeting around Grok Build's API rate should check xAI's current documentation directly rather than trusting either figure here.
··········
Grok Build: key figures
Detail | Value |
|---|---|
Original CLI launch | May 14, 2026, early beta |
Open-sourced | July 15, 2026, Apache 2.0 |
Build Mode (consumer) launch | July 28, 2026, SuperGrok Heavy only |
Expanded to all tiers incl. Free | By August 25, 2026 |
Context window | 256K (grok-build-0.1) or 500K (Grok 4.6) |
Parallel sub-agents | Up to 8, isolated Git worktrees |
Publishing | grok.me subdomain, custom domain, GitHub export |
··········
FREE ACCESS AND CURRENT LIMITS.
Grok Build went from a $300-a-month subscriber perk to something every account can try, in under a month.
As of August 25, 2026, xAI lists Grok Build as available on every plan, including Free — a sharp change from the SuperGrok Heavy exclusivity it launched with four weeks earlier.
SuperGrok, at $30 per month, and SuperGrok Plus, at $100 per month, raise usage limits rather than unlocking the feature itself, since it's already open on Free.
For unattended or headless use, an xAI API key bypasses the consumer app entirely and bills per token instead of drawing from a subscription's usage pool.
··········
Access by plan
Plan | Price | What it changes |
|---|---|---|
Free | $0 | Build Mode available; limits described dynamically, not published as a fixed number |
SuperGrok | $30/month | Higher usage limits across Build, Chat, Imagine, Voice |
SuperGrok Plus | $100/month | Significantly higher usage, priority access, faster replies |
API key | Pay per token | No subscription usage pool; billed directly on consumption |
··········
The limits themselves are the part xAI is deliberately vague about. Paid plans draw from one shared weekly usage pool spread across Chat, Imagine, Voice, and Build rather than separate per-feature quotas, and xAI states plainly that it does not publish one fixed request count that holds for every tier and every kind of task.
That means a table claiming an exact number of free Build Mode generations per day would be guessing. The only reliable way to see where an account actually stands is the usage view inside the account settings, not a number repeated from an older guide.
One practical consequence worth knowing: running several sub-agents in parallel draws down that shared weekly pool faster than a single terminal session would, so a parallel-agent workflow is the thing to budget around if usage is a concern, not the headline plan price.
··········
WHAT THIS MEANS FOR SUPERGROK HEAVY.
Bundling a full app builder into the $300 tier — and then giving the core feature away for free a month later — is a specific kind of business bet.
Before Build Mode, the strongest case for paying $300 a month was priority access to xAI's larger models and higher usage caps. Adding publishing and GitHub export briefly turned that tier into something closer to a developer-platform seat than a chatbot subscription.
Opening Build Mode to Free accounts within a month reframes that again: the $300 tier's case now rests on usage headroom and model priority rather than access to the feature itself.
That fast a walk-back from an exclusive launch is itself informative — it suggests the free-tier expansion was a deliberate distribution play once the feature had stabilized, not a sign the SuperGrok Heavy tier failed to justify itself on its own.
··········
HOW TO TRY IT.
Getting to a published app takes a handful of steps rather than a setup process.
Open Grok on web, iOS, or Android and switch to Build Mode from within the app.
Describe the app in a single prompt — the more specific the request, the closer the first draft lands to what's wanted, the same rule that applies to most agentic coding tools.
Review the plan the agent proposes before it starts executing, since that review step is where catching a wrong assumption is cheapest.
Once the build is working, publish to a grok.me subdomain for something quick, or connect a custom domain for anything meant to stick around — and export to GitHub if the plan is to keep developing it outside Grok.
··········
·····
FOLLOW US FOR MORE.
·····
·····
DATA STUDIOS
·····
