Grok Build 0.1: xAI's Agentic Coding API Open to Developers

The agentic coding market just got more crowded. On June 1, 2026, xAI announced that Grok Build 0.1 — previously accessible only to SuperGrok and X Premium+ subscribers via the CLI — is now available via the xAI API in public beta. That means any developer can call the model directly, integrate it into their own tooling, or build applications on top of the same model that powers the Grok Build terminal agent. At $1 per million input tokens, it enters the market at a price point that will force serious comparisons with Claude Code, OpenAI Codex CLI, and GitHub Copilot Workspaces.

This article covers everything: what Grok Build 0.1 actually does, how it's priced against competitors, the parallel agent architecture that sets it apart, where it falls short on benchmarks, and who should actually reach for it. We've verified all sources, benchmarks, and pricing figures cited below.

What Is Grok Build 0.1 and How Did We Get Here?

Grok Build launched as a terminal coding agent on May 14, 2026 — xAI's direct answer to Claude Code and OpenAI's Codex CLI. The underlying model, grok-build-0.1, was published separately on May 20, 2026, and entered API public beta on June 1. This separation matters: the CLI is a product; the API model is an infrastructure primitive that developers can use however they want.

The model was trained specifically for agentic software engineering — multi-step workflows where an AI needs to plan, search a codebase, write code, and verify results, not just produce a one-shot answer to a prompt. It accepts both text and image inputs (useful for dropping in screenshots of UI bugs or error messages), always has reasoning active, and natively supports function calling and structured outputs. xAI describes it as their fastest coding model, clocking over 100 tokens per second — a meaningful advantage for parallel workloads where latency compounds across multiple agents.

What Does Grok Build 0.1 Cost via the xAI API?

Pricing is straightforward: $1.00 per million input tokens and $2.00 per million output tokens, confirmed by xAI's official announcement and listed on OpenRouter. For context, a typical agentic coding session that reads 50,000 tokens of context and produces 5,000 tokens of code would cost approximately $0.06 — well under a cent for most practical tasks.

New xAI API accounts receive trial credits, making initial experimentation effectively free. Access the API at x.ai/api — no SuperGrok subscription required. Here's how the pricing stacks up against the main agentic coding competitors:

The price-performance trade-off is clear: Grok Build 0.1 sits 17 points below the top SWE-Bench scorers but costs 10-75x less per token. For teams running high-volume agentic pipelines where engineering depth isn't the constraint but cost and throughput are, that gap is very attractive. Explore more pricing breakdowns on our AI Free Tier Tracker.

How Does the Parallel Agent Architecture Work?

The headline differentiator for Grok Build is parallelism. According to codersera.com's detailed analysis, a single prompt can spawn up to eight sub-agents simultaneously, each working on an isolated Git worktree — meaning parallel edits cannot overwrite each other on the main branch. This is a concrete architectural advantage for large codebases where sequential single-agent passes are the bottleneck.

Each agent follows a three-stage workflow:

• Plan — the agent produces an explicit, step-by-step execution plan before touching a single file. In the CLI, you can approve the whole plan, comment on individual steps, or rewrite parts before any code is written.
• Search — the agent reads relevant files, navigates the repo structure, and builds its working memory about the codebase.
• Build — the agent executes the approved plan, runs commands, writes code, and verifies results against tests where available.

Arena Mode sits on top of this. When multiple agents produce competing solutions, Arena automatically scores them on automated evaluation criteria — do the tests pass? What's the diff size? Does the solution adhere to the approved plan? — and ranks them before they hit your review queue. Rather than drowning in eight raw outputs, you get a ranked shortlist. Arena is currently rolling out across the public beta and is not yet enabled by default in every install.

How Does Grok Build 0.1 Compare to Claude Code and Codex?

On raw benchmark numbers, Grok Build 0.1 trails the category leaders. As covered by BuildFastWithAI, Claude Opus 4.7 scores 80.4% on SWE-Bench Verified (other sources cite 87.6% on the updated version), benefits from a 1M token context window, and has more reliable tool-use traces in complex multi-file engineering scenarios. GPT-5.5 scores even higher at 88.7% but comes with a significantly higher price tag.

The New Stack's six-month comparison of agentic coding tools frames it clearly: Grok Build is best positioned for parallel-heavy migrations, high-volume API workloads, and multi-agent orchestration stacks — not as a first-choice agent for depth-sensitive, mission-critical engineering work. For teams already running Claude Code or Cursor, Grok Build makes more sense as a secondary model in a routing layer than as a replacement.

Where Grok Build does compete directly is on speed and price. At 100+ tokens per second and $1/$2 per million tokens, running eight agents in parallel on Grok Build costs roughly the same as running a single slower agent on a premium model. That arithmetic matters for automated pipelines, CI-triggered refactors, and any workflow where cost efficiency at scale is the primary constraint. See our AI model comparison pages for detailed side-by-sides.

What Integrations and MCP Support Does Grok Build 0.1 Offer?

One of the more strategically interesting aspects of Grok Build is its MCP posture. Rather than building a closed ecosystem, xAI adopted Model Context Protocol — the open standard championed by Anthropic — and implemented a "Bring Your Own MCP" approach. Internal knowledge bases, proprietary APIs, and custom MCP gateways can be plugged directly into Grok Build without modification. Critically, the system is explicitly compatible with Claude Code skills, plugins, and CLAUDE.md files, meaning teams already invested in that tooling can migrate or run both in parallel without rebuilding their skill packs.

The first wave of native Connectors launched May 6, 2026, covering GitHub, Notion, Linear, Google Workspace, and Microsoft 365. A May 22 expansion added Vercel, Canva, Gamma, and S&P Global. This is a broad enough integration surface to make Grok Build viable for real product teams, not just solo developers experimenting with a new CLI. For developers building on top of the API, your own AGENTS.md, hooks, and MCP servers load automatically when Grok Build starts in your repo.

If you're evaluating other open and accessible coding tools, our roundup of the best free AI coding assistants in June 2026 covers the broader landscape including fully free options.

Who Should Actually Use Grok Build 0.1 via the API?

The right fit for Grok Build 0.1 comes down to what you're optimizing for. Here are the clearest use cases based on the model's actual strengths:

• High-volume automated refactors — teams running large-scale migrations (e.g., upgrading a dependency across 200+ files) benefit most from the 8-agent parallel architecture at Grok Build's price point.
• Multi-agent orchestration layers — developers building orchestration systems who want to route simpler coding tasks to a fast, cheap model without saturating their Claude or GPT-5.5 budget.
• Prototyping and exploratory coding — the trial credits available to new xAI API accounts make it a near-zero-cost entry point for developers evaluating agentic coding workflows for the first time.
• Teams already using xAI infrastructure — if your stack already uses Grok 4 or Grok 4 Fast for general tasks, adding Grok Build 0.1 as the coding-specialized model is a natural extension with no new vendor onboarding.
• MCP-heavy workflows — projects with existing MCP infrastructure (Notion, GitHub, Linear, custom gateways) that want a coding agent without rebuilding integrations from scratch.

The use cases to avoid: production-critical engineering tasks where SWE-Bench accuracy in the high-80s percentage range is required, codebases larger than 256K tokens that need full-context reasoning, and teams that rely on Cursor's IDE-native experience rather than a terminal agent or API.

As DevOps.com reported on June 1, 2026, industry analysts see Grok Build as a model that will compete on price and parallelism rather than depth — "leaving Grok Build to compete on price and parallelism rather than as the agent of record." That framing is accurate, and also describes a genuinely valuable market position in an agentic pipeline world where not every task needs the most expensive model.