xAI launched Grok Build on May 14, its first-ever AI coding agent, and CEO Elon Musk personally posted the call for feedback on X as the tool is in early beta. xAI describes it as “an agentic CLI for coding, building apps, and automating workflows” and available exclusively to SuperGrok Heavy subscribers at $299 per month, though xAI is currently offering an introductory deal at $99 per month for the first six months.
This comes after Musk replied to a question on X about the product's release window in April by saying: "It will take until May to be close to Opus 4.6 and June to match and maybe exceed. Short time by normal standards, but long time in the AI arena." Bloomberg confirmed the launch marks xAI's first serious push into professional coding, a market where Anthropic's Claude Code has driven the company to $30 billion in annual recurring revenue as of April 2026, up from $14 billion just two months prior.
Looking at the competition, Claude Code and OpenAI's Codex both have head starts in the AI coding agent race. Codex has surpassed three million weekly active users. Claude Code is Anthropic's primary growth engine. Grok Build enters with none of that history, but with a set of bets, it is now asking developers to test. Here's what it does.
/1. Grok Build Runs Up to 8 AI Agents at the Same Time
Most AI coding agents work sequentially, one AI working through tasks step by step. Grok Build spawns up to eight concurrent specialized sub-agents that simultaneously plan, search documentation, and write code in parallel.
For large, multi-file projects, a complex task gets divided and attacked from multiple directions at once instead of one at a time. Each sub-agent can run inside its own separate branch of the codebase, so parallel lines of work don't overwrite each other. This is the core architectural bet xAI is making.
/2. Arena Mode: AI Agents Compete and Rank Their Own Output Before You See It
Arena Mode is Grok Build's most discussed feature. Rather than presenting a single solution for the developer to accept or reject, Arena Mode runs multiple agents against the same problem, has them compete, and then ranks their outputs before the result reaches the developer.
All agent responses appear side by side with a usage tracker, scored, and ordered. Arena Mode was confirmed in code traces in February 2026; it is not yet live in today's early beta, but xAI has confirmed it as a coming feature.
/3. Plan Mode Lets You Approve Every Step Before Any Code Changes
Grok Build includes a plan mode that addresses one of the most common frustrations with AI coding agents: the tool starts executing, goes in the wrong direction, and by the time you notice it has already rewritten a dozen files. In plan mode, Grok Build presents the full execution plan first.
You can approve the plan, comment on individual steps, or rewrite parts entirely before a single line of code is touched. Once a plan is approved, every change surfaces as a clean diff, a clear before-and-after view of what was changed. This gives developers more visibility and control upfront.
/4. Your Code Never Leaves Your Machine
Grok Build is built local-first: all code runs on your machine, and nothing in your codebase is transmitted to xAI's servers during a session. The tool is also air-gap compatible, meaning it works in sensitive offline environments once the initial setup is complete.
For developers working on proprietary codebases, contractors operating under NDAs, or teams in regulated industries like finance and healthcare, where sending source code externally is a compliance risk, this is a meaningful difference from tools that route operations through company servers.
/5. It Works With the Tools Developers Already Use, Including MCP Servers
Grok Build picks up existing developer conventions when launched inside a project folder. AGENTS.md files (instruction files that tell AI agents how to navigate a codebase), plugins, hooks, skills, and MCP servers (the connectors that let AI tools talk to external services) all work out of the box without additional setup.
It also integrates with VS Code for developers who prefer a graphical interface alongside the command line. A headless mode allows Grok Build to run inside scripts and automated workflows without any interactive interface. The command-line tool also includes full ACP support, meaning developers can build their own bots and agent orchestration systems on top of it.
/6. The Underlying Coding Model Costs $0.20 per Million Input Tokens
The model powering Grok Build, grok-code-fast-1, was built from scratch by xAI with a training dataset focused on programming content and real-world pull requests. It scored 70.8% on SWE-Bench Verified, the industry's standard coding benchmark, using xAI's internal testing setup, and carries a 256,000-token context window, meaning it can hold a large codebase in memory across an entire session.
It is priced at $0.20 per million input tokens and $1.50 per million output tokens via API, competitive against what developers pay for Claude Code or Codex at comparable task volume. For teams running high-frequency automated coding loops at scale, that per-token gap compounds quickly. xAI itself notes that SWE-Bench benchmarks "don't fully reflect the nuances of real-world software engineering," and independent replication of the 70.8% score has not yet been published.
