xAI on May 14 announced Grok Build, an early-beta coding agent and command-line interface for software engineers, entering a fast-moving sector that includes Anthropic’s Claude Code, OpenAI’s Codex and GitHub’s Copilot. The terminal-based tool runs on macOS, Linux and Windows via PowerShell, and ships with plan mode, MCP server support and the ability to dispatch parallel subagents.
The release marks xAI’s first dedicated coding agent and ties the tool directly to its existing subscription ecosystem—SuperGrok and X Premium Plus—shifting the lab’s ambition from consumer chatbots to daily engineering workflows. For the developers and enterprises already evaluating coding agents, Grok Build offers a new native option running on xAI’s own model family.
The CLI supports interactive text-user interface use and headless scripting, so it can be embedded in continuous-integration pipelines. It integrates with the AGENTS.md convention already used by tools like Claude Code, easing adoption. Users can install plugins, define hooks and add custom skills; the tool also supports the ACP integration protocol. The parallel subagent system is a standout: each subagent spawns as an independent child session in its own worktree, allowing multiple tasks to proceed concurrently. That design mirrors the multi-agent pattern that competing tools are adopting, though xAI has not disclosed a cap on the number of simultaneous subagents.
When news first broke on May 14, several outlets reported that Grok Build was available only to SuperGrok Heavy subscribers—a tier that some reports pegged at $300 per month, though that price could not be independently confirmed against xAI’s public pricing pages. By May 25, when xAI published its official news page "Introducing Grok Build," the eligibility had widened to all SuperGrok and X Premium Plus subscribers. The shift suggests either a rapid expansion of the beta or a clarification of messaging. Subscribers can install the tool via curl on macOS/Linux and PowerShell on Windows, authenticating through a browser session or a supplied XAI_API_KEY.
Under the hood, Grok Build runs on grok-build-0.1, a model also available in early access through the xAI API. It offers a 256,000-token context window and pricing of $1.00 per 1 million input tokens and $2.00 per 1 million output tokens. That positions it competitively on cost for large codebase contexts. The API availability means enterprises could, in theory, build custom tooling atop the model without using the CLI, though xAI has not disclosed any enterprise-specific deployment terms or data-retention policies.
The coding-agent category has become crowded, with Anthropic’s Claude Code, OpenAI’s Codex agent, GitHub Copilot Coding Agent and Google’s Gemini CLI all vying for developer mindshare. xAI’s entry is notable because it pairs a proprietary model with a terminal-native workflow that emphasizes plan-before-execution and developer-convention compatibility. It also locks Grok Build into xAI’s subscription stack, which could help retention but also limits the addressable market compared to standalone tools that work with third-party APIs.
Grok Build remains in early beta, and no independent performance benchmarks, user counts or enterprise adoption data have been released. The exact rate limits for each subscription tier are unclear. The relationship between the CLI command "grok" and the product naming is still evolving. Any claims about reshaping developer toolchains are premature. The reported $300 Heavy tier price remains unverified, and the current expansion may foreshadow further changes. The absence of known launch partners or case studies means the tool’s real-world reliability is untested.
For the executives and technical leaders assembling a coding-agent strategy, Grok Build represents a new, model-native entrant that is already integrated with the X platform’s subscription base. The move puts pressure on rivals to differentiate on terminal-native features and subscription bundling, while xAI will need to prove the tool’s dependability in production environments before it can move beyond early adopters.