xAI releases Grok Build 0.1, a fast model trained specifically for agentic coding, with an Agent Client Protocol for orchestration

xAI released Grok Build 0.1 on May 19, 2026, a model trained specifically for agentic coding tasks rather than general assistance. The release followed a May 14 beta launch of the Grok Build product, which includes an interactive TUI, headless scripting mode, and an Agent Client Protocol (ACP) for integrating the model into larger orchestration systems.

What's new

Grok Build 0.1

• Model slug: grok-build-0.1
• Described by xAI as a fast coding model trained specifically for agentic coding, currently in early access
• Not a general assistant: purpose-built for agentic software engineering workflows
• Available in early access via the xAI API

Grok Build product (May 14 beta)

• Interactive TUI for developers using Grok Build directly in the terminal
• Headless mode: runs Grok Build in scripts and CI pipelines without a UI
• Agent Client Protocol (ACP): structured interface for integrating Grok Build into apps, orchestrators, and multi-agent systems
• No SDK wrapping required — ACP communicates directly over a defined protocol

Context

Grok Build follows a pattern now visible across multiple frontier labs: dedicated coding-agent models that sacrifice breadth to deepen software engineering performance. Anthropic has Claude Code (backed by the Opus models). GitHub Copilot Workspace uses OpenAI models tuned for coding. Google ships Gemini in IDEs with coding-specific prompting. The thesis across all of them is that a model fine-tuned end-to-end on agentic coding tasks produces better results for that workflow than a general model with a coding prompt.

The Agent Client Protocol is xAI's most distinctive contribution here. Rather than shipping just an API endpoint, xAI is defining a protocol layer — a way for Grok Build to be a first-class node in a multi-agent graph rather than a standalone tool. The ACP means that orchestrators built on other platforms can route coding subtasks to Grok Build without rebuilding integrations from scratch.

Timing is notable: Grok Build 0.1 launched the same day Google shipped Gemini 3.5 Flash GA and Managed Agents. It followed Anthropic's Claude Opus 4.8 by nine days.

Why it matters

Specialized coding models are winning over general models for agentic software engineering tasks in head-to-head evals. xAI entering this space with a model trained specifically for the agentic coding loop — rather than adapting a general model — signals that the lab sees purpose-built models as necessary to compete.

The headless mode and ACP are the production-readiness signals. Interactive TUIs are for developer adoption; headless scripts and orchestration protocols are for production pipelines. xAI is shipping both at once, suggesting Grok Build is positioned for enterprise developer tooling from launch rather than as a consumer product.

Early access limits who can evaluate Grok Build 0.1 today, but the ACP's existence means that when access broadens, integrating Grok Build into existing agent architectures will be straightforward. That lowers the switching cost — and raises the stakes for incumbents.