xAI Launches Grok Build in Early Beta to Take On Claude Code and Codex CLI

xAI has launched Grok Build, an agentic command-line coding tool for professional software engineering. The early beta is live now for SuperGrok Heavy subscribers, putting xAI directly against Anthropic's Claude Code and OpenAI's Codex CLI.

How Grok Build Works

Grok Build runs in the terminal. It plans, writes files, executes shell commands, and builds applications from natural language prompts. It is not a chat assistant. It is a coding agent that executes tasks end-to-end.

Core features:

Plan Mode: Proposes a full execution plan before touching any file. Developers approve, edit individual steps, or rewrite the plan before any code runs
Parallel sub-agents: Spawns up to 8 concurrent AI agents working simultaneously on separate codebase branches
Headless mode: Runs agents inside scripts and automations via the p flag
ACP support: Full Agent Client Protocol support for building bots and multi-agent orchestration
Zero-setup compatibility: Reads existing AGENTS.md, plugins, hooks, skills, and MCP servers directly from the repo

Install command: curl -fsSL https://x.ai/cli/install.sh | bash

Model Specs and Benchmark Results

Grok Build runs on Grok 4.3 beta, built on xAI's 16-agent Heavy architecture.

Context window: 256,000 tokens
SWE-Bench Verified score: 70.8%
API pricing: $0.20 per million input tokens, $1.50 per million output tokens

Who Can Access It and What It Costs

Access is restricted to SuperGrok Heavy subscribers at $299 per month. xAI has introduced an introductory SuperHeavy tier at $99 per month for the first six months.

Platform support at launch:

macOS: native
Linux: native
Windows: WSL2 only. A native Win32 build is on the roadmap, with no release date announced

The Market Context Behind the Launch

xAI CEO Elon Musk previously admitted the company fell behind rivals in coding. Musk called publicly for beta testers on X on the day of the launch.

The release comes months after xAI was acquired by SpaceX in February 2026. The company has since lost more than 50 researchers and engineers. Grok Build is the most technically substantial product shipped since that merger.

The enterprise coding agent market was a two-horse race between Claude Code and Codex CLI before May 14. Developers can compare Grok 4.3 and other leading AI models across real engineering tasks.

Grok Build vs. Claude Code vs. Codex CLI

Three features separate Grok Build from the incumbents:

Plan Mode graph view: Claude Code and Codex CLI produce linear text plans. Grok Build generates a graph of sub-tasks with per-node state in a dedicated terminal UI
Native parallel sub-agents: Up to 8 agents run simultaneously on independent branches. Neither Claude Code nor Codex CLI ships native parallel execution at this level
ACP-first architecture: Built on the open Agent Client Protocol from the ground up for multi-agent orchestration without a chat layer

Grok Build reads the Anthropic Skills format and supports MCP servers out of the box. Claude Code users can bring existing tooling over without modification.