xAI launched Grok Build on May 15, an agentic coding CLI that runs in your terminal and competes directly with Claude Code and OpenAI Codex. The initial beta is gated to SuperGrok Heavy subscribers at $300 per month, with an introductory deal of $99 per month for the first six months. The backend is Grok 4.3 beta, running a 16-agent Heavy architecture and a 2-million-token context window.

How to Try It

SuperGrok Heavy users log into x.ai, download the beta installer, and run grok build in a project directory. The CLI opens with a Plan Mode by default: it proposes a multi-step plan, lets you approve or rewrite individual steps, and only then executes each step as a reviewable diff. For larger tasks it dispatches subagents in parallel and merges their output back into the trunk plan. Cancel any step with Esc, or reroute it with an inline comment before approval. The CLI ships for macOS and Linux today, with a Windows build in beta.

What Happened

According to basenor's launch coverage, Grok Build inherits xAI's existing Heavy multi-agent runtime, which has previously been exposed only through the Grok chat interface. The 16-agent setup is the same one that posts on Artificial Analysis benchmarks for reasoning. Putting it behind a CLI is xAI's first developer-product release of 2026, and the rollout follows a familiar pattern: launch on the top SuperGrok tier first, then trickle down to lower tiers within 30 to 60 days.

Why It Matters

This is the third top-tier lab to ship a terminal-native coding agent in six months, after Claude Code and OpenAI Codex CLI. Grok Build's pitch is the 2-million-token context, which lets you load entire monorepos at once instead of chunking. The parallel 16-agent dispatch is the other differentiator: Claude Code runs serial subagents and Codex CLI runs one task at a time. If those two claims hold on real codebases, Grok Build slots in for refactor-heavy work where the agent needs both wide context and parallel reach.

The price is the catch. According to Kingy AI's pricing breakdown, $300 per month is roughly 5x Claude Pro and 3x ChatGPT Pro. Even with the $99 intro deal, the first year of access costs $1,188. Solo developers will probably wait for the cheaper-tier rollout; teams running large monorepos and treating coding agents as paid headcount can already justify it.

Key Details

  • Launch date: May 15, 2026
  • Tier: SuperGrok Heavy ($300/mo, or $99/mo for the first 6 months)
  • Backend: Grok 4.3 beta, 16-agent Heavy architecture, 2M token context
  • Interface: terminal CLI with Plan Mode, diff review, parallel subagents
  • Platforms: macOS and Linux at launch, Windows in beta
  • Distribution: SuperGrok account login, beta installer from x.ai

What to Do Next

If you already pay SuperGrok Heavy, the beta is a free upgrade. Install it tonight and run it on a side project to see whether Plan Mode catches more before it codes than Claude Code's serial agent does. If you do not pay SuperGrok Heavy, wait for the cheaper-tier rollout. For teams: benchmark Grok Build against Claude Code on your own monorepo before switching, especially on the 2M context claim. The first month of community benchmarks will tell you whether the parallel dispatch holds outside xAI's curated demos.