xAI releases Grok Build: an agentic coding CLI and grok-build-0.1 API model with plan mode, parallel subagents, and MCP support

xAI launched Grok Build in May 2026, entering the agentic coding tool space alongside Anthropic's Claude Code, OpenAI's Codex CLI, and GitHub Copilot Workspace. The release packages two connected components: a terminal CLI with an interactive interface and headless scripting mode, and a dedicated API model — grok-build-0.1 — purpose-built for software engineering agents.

What's new

"Grok Build is now available in beta. Use the interactive TUI, run headlessly in scripts, or build apps and orchestrators with the Agent Client Protocol."

Grok Build CLI is a terminal coding agent installable via a single shell command. Its headline features:

• Plan mode: Generates a graph of sub-tasks with per-node state tracking displayed in the TUI
• Parallel subagents: Independent agents work on separate code branches simultaneously
• Git worktree support: Isolates branch changes so concurrent work stays clean
• Agent Client Protocol (ACP): Lets developers build multi-agent orchestrators on top of Grok Build
• MCP server support: Extends the CLI with custom tools via the Model Context Protocol
• Headless mode (-p flag): Runs agents inside CI scripts and automations without a UI

Initial beta access was gated to SuperGrok Heavy subscribers, with broader public beta availability opening in late May.

grok-build-0.1 is the underlying API model, also accessible directly through the xAI API for developers who want to build their own tooling. It runs at over 100 tokens per second, supports a 256K context window, and is priced at $1 per million input tokens and $2 per million output tokens. The model is described as specifically trained for agentic coding tasks including web development, debugging, and tool use.

Context

The agentic coding CLI category has compressed rapidly in 2026. Anthropic's Claude Code launched in late 2025 and reached general availability in early 2026. OpenAI's Codex CLI went multi-model in spring 2026. Cursor and Replit have built product moats around AI-assisted development. xAI enters with a throughput claim — 100+ tokens per second — positioned as a differentiator for interactive use cases where generation latency directly affects how fast a developer can iterate.

The underlying model is built on xAI's Grok 4.3 architecture, the same 16-agent Heavy system the company uses for its Grok flagship models. The ACP integration means Grok Build can serve as a node in larger multi-agent pipelines, not just a standalone terminal assistant.

Why it matters

xAI's reputation has centered on Grok's speed and its web-connected inference. Grok Build is the company's first product that competes directly for developer workflow time — the same ground Anthropic and OpenAI are fighting over with Claude Code and Codex. The pricing ($1/$2 per MTok) places grok-build-0.1 well below Claude Sonnet tiers and competitive with GPT-4o on a per-token basis. For teams already paying for SuperGrok Heavy, the CLI arrives as a bundled benefit. For developers building agents, ACP and MCP support make Grok Build a composable piece in orchestration stacks rather than a walled terminal experience. How it performs against Claude Code on real engineering tasks will be the practical test, but the architecture — plan graphs, parallel subagents, worktree isolation — maps closely to what Anthropic shipped at launch.