The coding agent market is shaking up again. On May 14, 2026, xAI launched Grok Build in beta — becoming the third entrant in the terminal coding agent race alongside Claude Code and Codex CLI.
What's interesting is xAI's strategy. Instead of chasing benchmark scores, they bet on a structurally different approach — 8 agents coding simultaneously, each on independent branches. Here's whether that bet pays off.
So what exactly is this thing?
Until mid-2025, terminal coding agents were effectively a two-horse race: Anthropic's Claude Code and OpenAI's Codex CLI. xAI's Grok was widely acknowledged to lag behind in coding capabilities — Elon Musk himself admitted it.
Grok Build is xAI's direct response to that gap. Rather than adapting a general-purpose model for coding, they built grok-build-0.1 from scratch for agentic workflows — trained specifically for multi-step code execution, not repurposed from general chat.
It launched May 14 as a SuperGrok Heavy ($299/month) exclusive, then expanded May 25 to all SuperGrok ($30/month) and X Premium+ subscribers.
The local-first architecture is worth noting. Source code never gets sent to xAI's servers — it works in air-gapped environments too. That's a meaningful unlock for financial services, healthcare, and government teams where Claude Code or Codex CLI can't go.
Native MCP (Model Context Protocol) support means existing Claude Code integrations — GitHub, Linear, Slack — carry over with zero reconfiguration.
What's actually different from Claude Code?
On paper, Grok Build is behind. SWE-Bench Verified: Claude Code at 87.6%, Codex CLI at 88.7%, Grok Build at 70.8% — a 17-point gap. That's not a rounding error. It shows up in complex, multi-file tasks requiring deep reasoning.
But xAI is playing a different game. The bet isn't benchmark performance — it's how the work gets done.
| Claude Code | Codex CLI | Grok Build | |
|---|---|---|---|
| SWE-Bench | 87.6% | 88.7% | 70.8% |
| Parallel agents | Supported | Supported | Up to 8 (default) |
| Plan approval | Optional | Not available | On by default |
| Local-first | No | No | Yes (air-gap) |
| MCP support | Native | Not supported | Native |
| Entry price | $20/mo (Pro) | $20/mo (ChatGPT+) | $30/mo (SuperGrok) |
Two differentiators worth understanding:
Plan Mode on by default: Grok Build writes a plan before touching any code. You review, edit, and approve before execution starts. Claude Code has Plan Mode as an option; Codex CLI doesn't have it at all. This three-step gate (plan → review → execute) structurally prevents the agent from running off in the wrong direction.
Real parallelism: 8 agents each work on isolated Git branches simultaneously. For a legacy auth module refactor, one agent handles the core logic, another writes tests, a third updates docs — at the same time. Reviewers report complete CRUD API generation with auth and tests in around 15 minutes.
Where Grok Build actually shines
Large monorepo refactoring, bug investigations requiring parallel hypothesis testing, and architecture audits — anywhere you need to explore multiple directions simultaneously. For simple feature additions or 1:1 debugging, Claude Code and Codex CLI are more reliable.
Quick start: how to get going
- Install
Run the official install script. macOS/Linux supported; Windows needs WSL2.curl -fsSL https://x.ai/cli/install.sh | bash - Login
Authenticate with your SuperGrok or X Premium+ account.grok-build login - Start with plan mode
Run with --plan flag — it shows you the plan first. Review, adjust, then approve to start execution.grok-build --plan "refactor auth module" - Add parallel agents
Use --parallel for complex tasks. Start with 2-4, scale up as you get comfortable.grok-build --parallel=4 "task description" - Bring your MCP setup
Existing Claude Code MCP configurations work out of the box. GitHub, Linear, Slack integrations carry over without reconfiguration.
Watch the pricing tiers
The $30/month SuperGrok plan gives basic access. Full parallel agents and Arena Mode require SuperGrok Heavy ($99/month for 6 months introductory, then $299/month). API-only pricing: $0.20 per million input tokens, $1.50 per million output tokens.
Go deeper
ChatForest: Grok Build vs Claude Code vs Codex CLI Deep Review The most comprehensive benchmark comparison including real PR generation tests chatforest.com
Codersera: Decision Matrix by Use Case Scenario-by-scenario breakdown of which agent to pick for your workflow codersera.com
ByteIota: Honest Grok Build Review Balanced developer perspective covering strengths and real limitations byteiota.com
CIO Dive: xAI Enters the Coding Agent Race Enterprise perspective on market positioning and competitive dynamics ciodive.com
Jingrey: Grok Build Beta In Practice Real development task tests including legacy auth module refactoring jingrey.com
Engadget: Grok Build Launch Coverage Official announcement and market context engadget.com
