Grok Build: xAI's Terminal Agent on a Subscription Meter

Grok Build is xAI's terminal coding agent, launched in early beta on May 14 2026 for SuperGrok Heavy subscribers and opened to all SuperGrok (\$30/mo) and X Premium+ (\$40/mo) subscribers on May 25 2026. You install it with one command (curl -fsSL https://x.ai/cli/install.sh | bash on macOS/Linux, irm https://x.ai/cli/install.ps1 | iex on Windows), run grok in a project directory, and it plans, edits, and validates multi-file changes. Its distinguishing features are up to 8 parallel subagents each isolated in its own git worktree, plan mode on by default (nothing touches your files until you approve the plan), headless -p mode for scripts and CI with streaming JSON output, Agent Client Protocol support for editors, and out-of-the-box compatibility with the Claude Code ecosystem - AGENTS.md, skills, hooks, and MCP servers load unchanged. The Rust source is public at xai-org/grok-build (July 2026).

There is no standalone Grok Build plan - access is bundled with xAI subscriptions: SuperGrok at \$30/mo, X Premium+ at \$40/mo, or SuperGrok Heavy at \$300/mo for the highest limits. If you already pay for SuperGrok or X Premium+, the agent is free upside on a plan you have. Third-party reviewers also reported a limited free tier of about 25 parallel agent runs per day (June 2026) - it is not documented on xAI's own pages, so verify what your account actually shows. The no-subscription path is the API: grok-build-0.1 at \$1 per 1M input and \$2 per 1M output, pay as you go.

The cheapest bundled path is SuperGrok at \$30/mo - the lowest first-party terminal-agent subscription in the field. X Premium+ costs \$40/mo with the same agent access plus X platform perks, and SuperGrok Heavy costs \$300/mo with the highest limits. There is no mid-tier between \$40 and \$300, so if SuperGrok's limits are not enough the next step is steep. Compare Claude Code Pro at \$20/mo and Codex included with ChatGPT Plus at \$20. On the API, budget math: agent loops re-send large prompts every turn, so cached input at \$0.20 per 1M is where the real savings live; below roughly 85 agent turns per day, metered API billing beats the \$30 subscription.

grok-build-0.1 is the coding model that powers the Grok Build CLI, released on the xAI API in public beta on May 29 2026 (formerly grok-code-fast-1). It scored 70.8% on SWE-bench Verified - solid but roughly 17 points behind the Claude Code and Codex frontier leaders - and trades context for cost and latency: 256K tokens of context (the smallest in the frontier-CLI field; Grok 4 itself carries 2M), 100+ output tokens per second, and pricing of \$1 per 1M input / \$2 per 1M output / \$0.20 per 1M cached input. Default rate limits are 1,800 requests per minute and 10M tokens per minute. The same model also runs in Cursor, Kilo Code, OpenCode, and any OpenAI-compatible agent harness.

For larger tasks Grok Build delegates work to up to 8 specialized subagents that run in parallel, each inside its own isolated git worktree, so parallel edits never collide and changes merge clean - the most distinct architecture decision in the terminal-agent category, and something Claude Code and Codex CLI do not ship (their subagents spawn per task with worktree management left to you). Each subagent follows a three-stage plan-search-build flow. Plan mode is on by default: Grok Build proposes a numbered plan, you approve, comment, or rewrite it, and only then does execution begin - every change surfaces as a clean diff. Deep worktree integration also lets you launch subagents in worktrees you choose.

Yes, natively - no WSL required. Windows installs with one PowerShell command (irm https://x.ai/cli/install.ps1 | iex) and config lands in %USERPROFILE%\.grok\config.toml; macOS and Linux use the curl installer. Interactive sessions authenticate through a browser, and headless or CI usage sets XAI_API_KEY. One caveat for contributors rather than users: the open Rust tree's Windows builds are best-effort and not tested by the maintainers - the shipped installer is the supported Windows path. Run grok inspect after setup to confirm which config sources, skills, plugins, hooks, and MCP servers the agent discovered in your project.

They are different meters for different jobs. Quality: grok-build-0.1 scores 70.8% on SWE-bench Verified against roughly 87-89% for Claude Code - about 17 points behind, and hands-on reviews consistently call Grok Build the value pick for scaffolding and bulk generation rather than the reliability pick for long autonomous builds. Price: \$30/mo via SuperGrok vs Claude Code Pro at \$20/mo (or Max at \$100-\$200). Context: 256K vs an effective 1M. Architecture: Grok Build's 8 worktree-isolated parallel subagents and default plan mode vs Claude Code's per-task subagents. Ecosystem: both read AGENTS.md and MCP servers, so migration is cheap either way. Frontier quality on complex refactors: Claude Code. Cheapest first-party parallel agent, especially if you already pay for X or SuperGrok: Grok Build.

Yes. VibeFuse is the first free widget-based AI harness: it runs the official Grok Build CLI - alongside Claude Code, Codex CLI, Gemini CLI, Cursor, and any Model-API agent - as widgets over your real repositories on Windows. The harness costs \$0 with no usage meter: your xAI subscription or API key does the billing, and you switch models or vendors per task instead of being locked to one meter. When a quota wall or rate limit hits one widget, route the task to another vendor's CLI on the same canvas. VibeFuse also runs an open-source marketplace where creators sell widgets, skills, and styling packs for revenue share - pair it with VocalFuse for local voice dictation while you build.