Grok Build launches beta CLI with plan mode, subagents, and $300 Heavy access

• xAI has started an early beta for Grok Build, a terminal coding agent that koltregaskes' launch summary and bridgemindai's screenshot both describe as a CLI for coding, app building, and workflow automation.
• The launch feature set centers on plan mode, approvals and diffs, skills or plugins, hooks, and parallel subagents in separate worktrees, according to koltregaskes' feature rundown and testingcatalog's beta post.
• The beta is gated behind SuperGrok Heavy, which koltregaskes' pricing screenshot prices at $300 per month, although testingcatalog's follow-up pointed to a $99-for-six-months offer page via this Grok pricing link.
• Early hands-on posts focused as much on the packaging as the product: bridgemindai's install failure hit a 404 on the installer, while haider1's UI screenshot and theo's promo grab highlighted the terminal interface and the "fast and flicker-free" pitch.

You can grab the one-line installer from bridgemindai's screenshot, see xAI pitch a "fast and flicker-free" CLI, and watch it running inside Emdash's demo. The more interesting bits are the familiar coding-agent primitives xAI bundled on day one, especially separate-worktree subagents, an ask-user-question tool, and headless mode for scripting and CI/CD.

Grok Build enters the same terminal lane as Claude Code, Codex, and Gemini CLI, with xAI framing it as an agentic CLI for professional software work. bridgemindai called it another player in the coding-agent race, but the concrete story is the mechanics xAI exposed immediately.

The launch bundle, as described across koltregaskes' post, testingcatalog's post, and mattlam_'s feature list, includes:

• Plan mode
• User approvals and clean diffs
• Skills
• Plugins
• Hooks
• AGENTS.md pickup
• Parallel subagents
• Separate worktrees
• An ask-user-question tool
• Headless mode
• ACP support for custom bots

The UI is doing some of the launch work here. haider1's screenshot shows inline task progress, edit traces, slash commands like /btw, and an always-approve mode in the status bar.

The subagent view in theo's screenshot and mattlam_'s screenshot is the sharper reveal: multiple agent lanes run in parallel, each with its own task label and timer, while the main session asks clarifying questions before kicking off more work. That puts Grok Build closer to the recent delegation-first agent CLIs than to a single-threaded chat shell.

The beta is not broadly available. koltregaskes' screenshot says "Exclusive access to Grok Build" sits inside the SuperGrok Heavy tier at $300 per month, with a three-day free trial on the pricing page.

That price dominated a lot of the immediate reaction. LLMpsycho's post and LLMpsycho's follow-up both fixated on the monthly cost, while testingcatalog's follow-up linked a discounted SuperGrok offer page advertising $99 per month for six months.

The paywall matters because it narrows the pool of real testers on day one. For now, the public evidence is mostly screenshots, short demos, and first impressions rather than broad hands-on comparison data.

The roughest early report was basic setup. bridgemindai's install screenshot shows the published curl ... | bash flow returning a 404 while trying to fetch grok-0.1.210-macos-x86_64 from xAI's public artifact bucket.

At the same time, xAI and early users were already pushing it into other shells and environments. Emdash's clip shows Grok Build running inside Emdash, and koltregaskes' pricing screenshot lists headless mode for scripting and CI/CD as part of the Heavy package. Those two details suggest xAI is shipping Grok Build as more than a branded terminal app, even if the first beta still looks rough around installation.