Grok Build: xAI's Terminal Coding Agent Under the Microscope

• Up to 8 parallel sub-agents that can plan, search, write code, and debug concurrently
• Plan-first execution loop: inspects the repository, proposes changes, and shows diffs for human review before applying anything
• 256K token context window for large-scale workflows
• Multiple interfaces: interactive TUI, headless mode for CI/CD, and Agent Client Protocol (ACP) for integration into other apps
• MCP (Model Context Protocol) compatibility, and it reads Claude Code config files (CLAUDE.md, skill packs) without modification
• /goal autonomous mode: long-running execution with pause/resume controls
• SWE-Bench Verified score of 70.8%
• Pricing: the underlying model is priced at $0.20 per million input tokens on the API ; the API public beta for grok-build-0.1 launched at $1/M input and $2/M output tokens

The agent follows a three-stage workflow — Plan, Search, Build — and runs sub-agents in isolated Git worktrees to prevent conflicts .

The privacy controversies that preceded the release

Before Grok Build launched, a series of overlapping privacy scandals had already drawn global regulatory scrutiny to xAI:

• Sexual deepfake scandal: From 2025 onward, Grok's image-generation capabilities were used to create non-consensual sexually explicit images of real people, including minors. Millions of such images were generated and publicly disclosed .
• Regulatory investigations: The UK's ICO opened a formal inquiry in February 2026 ; Canada's Privacy Commissioner found X and xAI violated PIPEDA in June 2026 ; the European Commission ordered X to retain all Grok documents through end of 2026 .
• Default data training: X enabled data-sharing for Grok training by default, drawing criticism from privacy advocates. Ireland's Data Protection Commission used emergency powers for the first time to stop X from using EU users' data .
• 370,000 leaked private chats: An estimated 370,000 private Grok chats were accidentally published and indexed publicly, including sensitive content .
• FTC petition: In June 2026, X petitioned the FTC to end the Twitter privacy consent decree, with critics arguing this would reduce data protections for Grok training .

These events created a tense backdrop for the Grok Build release, with regulators and privacy groups closely watching xAI's data practices.

How Grok Build fits into Musk's open-source pledges — a mixed record

Elon Musk has made a series of open-source promises for his companies, with mixed delivery:

The pattern: Musk has followed through on open-sourcing older models (Grok 2.5) as a goodwill gesture, but has repeatedly delayed or not honored promises to open-source newer, more commercially valuable products (Grok 3) . The Grok Build launch as a $300/month closed beta is consistent with this approach — it is a premium, revenue-generating product, not an open-source one. Meanwhile, the newer open-source pledge for X's codebase is separate and was announced the same day an article noted that Grok Build may have made exposing code easier .

The bottom line

Grok Build is a serious technical product — its parallel-agent architecture, Claude Code compatibility, and SWE-Bench score are competitive. But developers should not expect its code or model weights to be released publicly. The broader open-source rhetoric continues to generate goodwill and regulatory cover for xAI, but the actual track record on newer releases remains incomplete.