Grok Build: xAI's $300 Terminal Coding Agent

The Breakthrough

xAI launched Grok Build, a terminal-based coding agent that utilizes the Grok 4.3 model and achieves interoperability with existing Claude Code configurations by supporting the same agents.md files and MCP servers.

What Actually Worked

• Multimodal Refinement: The agent successfully identified and corrected visual rendering issues in a C++ game by analyzing screenshots provided directly into the terminal, demonstrating effective visual feedback loops.
• Native MCP Integration: Unlike competitors that require plugin-based bolt-ons, Grok Build integrates the Agent Client Protocol natively, allowing it to orchestrate other coding agents as worker processes.
• Parallel Execution: The system supports up to eight sub-agents running in parallel, which provides significant speed improvements for large-scale refactoring tasks.
• Memory Management: The tool includes a persistent memory subsystem featuring dream, search, edit, and flush commands to consolidate context across different coding sessions.

Before / After

• Model Consolidation: xAI retired four legacy models (Grok 4, Grok 4 Fast, Grok 4.1 Fast, and Grok Code Fast 1) five days prior to the launch, routing all traffic through the new Grok 4.3 model.
• Pricing: The service is gated behind the Super Grok Heavy tier at $300 per month, with a temporary promotional price of $99 for the first six months.

Context

xAI is attempting to capture the developer market by prioritizing compatibility with established workflows rather than competing solely on raw coding benchmarks. By adopting the agents.md standard and MCP support, they allow users to migrate from Claude Code with minimal friction. However, early testing reveals significant stability issues, including a broken plan mode that fails to trigger consistently and a sub-agent routing layer that occasionally pulls context from unrelated previous projects. The product is currently in an early beta phase, with xAI actively soliciting bug reports via an in-CLI feedback command.

Notable Quotes

• "The marketing says plan mode is the differentiator. The reality is plan mode is broken in a way that is pretty embarrassing for an early beta at this price."
• "The eight agent fan out works but the routing layer between them is not yet reliable."

Content References

{"type": "tool", "title": "Claude Code", "publisher": "Anthropic", "context": "mentioned"}, {"type": "tool", "title": "Codex CLI", "publisher": "OpenAI", "context": "mentioned"}