By ByteBot
xAI shipped Grok Build on May 14 — a terminal-native coding agent that drops directly into developer workflows alongside Claude Code and OpenAI’s Codex CLI. The three-way agentic CLI race is now official. Grok Build brings a 2 million token context window, a pre-execution Plan Mode, and up to 8 parallel sub-agents. Whether that justifies $99 to $300 per month is a harder question, and we’ll answer it honestly.
What Grok Build Actually Does
Grok Build is a command-line agent built on grok-code-fast-1, a model xAI built from scratch — not a variant of Grok 4 — with a training corpus heavy on programming content and post-training focused on real-world pull requests. It can plan projects, write and edit files, execute shell commands, and build complete applications from natural language prompts.
The feature that will make Claude Code users pay attention: Plan Mode. Before Grok Build touches a single file, it presents a full step-by-step execution plan. You approve the plan, comment on specific steps, or rewrite parts entirely. Only after sign-off does code execute — every change surfaces as a clean diff. This is the number-one feature request from Claude Code users, and xAI ships it by default. Six key differentiators from the Grok Build early beta are covered in depth by Techloy.
The 2 Million Token Context Window (When It Actually Matters)
Grok Build holds 2 million tokens in context — ten times Claude Code’s 200K. For most developers working on projects under 200K tokens, this advantage is invisible. But for teams managing large monorepos where context becomes the actual bottleneck mid-refactor, the gap is real. On benchmarks, grok-code-fast-1 scores 70.8% on SWE-Bench Verified — just below Claude Sonnet 4.6 at 72.7%, and well behind Claude Opus 4.7 at 87.6%. Speed is a genuine differentiator: the model runs at up to 160 tokens per second. The complete Grok Build CLI guide breaks down the architecture further.
Installation and Setup
Installation is one command:
curl -fsSL https://x.ai/cli/install.sh | bashmacOS and Linux are natively supported. Windows requires WSL2 — native Win32 support is on the roadmap with no announced date, which is a meaningful friction point for Windows developers. Once installed, Grok Build automatically picks up your existing AGENTS.md files, MCP servers, plugins, and hooks — including Skills in Anthropic’s format. Zero migration required from Claude Code. Headless mode for CI/CD automation:
grok -p "add integration tests for the auth module"Pricing: The Honest Breakdown
Access is currently limited to SuperGrok Heavy subscribers. The introductory price is $99/month for the first six months (a 67% discount from the regular $300/month). Here is how it compares:
| Tool | Cost |
|---|---|
| Claude Code Pro | $20/month |
| Claude Code Max | $100/month |
| Grok Build (intro) | $99/month (first 6 months) |
| Grok Build (regular) | $300/month |
At $99 intro, Grok Build sits competitively against Claude Max. At $300 regular — after the promotional period ends — this is a professional engineering team purchase. One developer on the Hacker News discussion thread put it plainly: “$300 a month on something my employer will never approve.” For solo developers, the intro window is your evaluation window.
Who Should Actually Try It Now
Grok Build is early beta with acknowledged incomplete error handling and subagent coordination issues. The honest answer: a specific subset of developers should evaluate it now.
Try Grok Build if: your codebase regularly exceeds 200K tokens, you’ve been frustrated by agents starting execution before confirming task comprehension, or you’re evaluating options for a professional team with budget for $300/month tooling. The local-first architecture — nothing from your codebase is transmitted to xAI servers during a session — also matters for teams in regulated industries.
Stay on Claude Code if: you’re a solo developer on the $20/month plan, you need Windows native support, or you need production-grade stability in CI/CD pipelines. The indie hacker perspective on the Claude Code vs Grok Build decision aligns with this read: Claude Code and Codex CLI remain the safer choices for production automation until Grok Build reaches GA.
What to Watch: Arena Mode
The feature that could change the calculus is Arena Mode — multiple agents competing against the same problem, with outputs ranked before human review. xAI confirmed it’s coming; it appeared in code traces back in February 2026. It is not live yet. If it ships well, it would add an automated evaluation layer that neither Claude Code nor Codex CLI currently offers.
Grok Build’s arrival is good news regardless of whether you switch. Competition accelerates feature development, and Plan Mode alone raises the bar for what a capable coding agent should ship with. Read xAI’s official Grok Build announcement for the full technical spec. Claude Code’s next release will be more interesting because Grok Build exists.
