Grok Build 0.1 Is Live on the xAI API at $1/$2 Per Million Tokens - Setup Guide for OpenRouter and Vercel

API vs CLI - Two Different Products Built on the Same Model

This is a common point of confusion worth clarifying upfront. Grok Build (the CLI app) is xAI's terminal-based coding agent — a TUI tool available to SuperGrok ($30/month) and X Premium+ subscribers. It launched in beta on May 25, 2026 and is what you use interactively to work on your own codebase. grok-build-0.1 (the API model) is the underlying model that powers the CLI, now available for developers to call directly via API — meaning you can embed it in your own applications, pipelines, and agent frameworks without using the Grok Build CLI at all.

The underlying model was released on May 20, 2026. API availability was confirmed officially on May 27 via xAI's news page and simultaneously on OpenRouter and Vercel AI Gateway. The xAI release notes confirm the model slug is grok-build-0.1 and that it "performs best in agentic harnesses like Grok Build, Cursor, Hermes Agent, OpenClaw, Kilo Code, or OpenCode" — giving you a direct list of the agent frameworks where it has been tested and performs well.

Model Specs and Capabilities

The always-on reasoning is an important implementation detail. Unlike Grok 4.3 where reasoning effort is configurable (none, low, medium, high), grok-build-0.1 always runs reasoning before producing its final output. Every response includes structured analysis before the answer. This means slightly higher latency per request than a non-reasoning model, but more reliable results on complex multi-step coding tasks. It is not adjustable — you cannot turn it off to save tokens on simpler tasks.

The no-output-token-cap is a practical advantage for long autonomous coding sessions that generate large amounts of code. Models with output caps (many enforce 4K or 8K limits by default) require workarounds in agentic pipelines to handle tasks that produce large outputs. grok-build-0.1 eliminates that constraint.

How to Access It on Each Platform

curl https://api.x.ai/v1/responses   -H "Content-Type: application/json"   -H "Authorization: Bearer $XAI_API_KEY"   -d '{
    "model": "grok-build-0.1",
    "input": [{
      "role": "user",
      "content": "Refactor this Python function to be async"
    }]
  }'

from openai import OpenAI

client = OpenAI(
    base_url="https://openrouter.ai/api/v1",
    api_key="your-openrouter-key"
)

response = client.chat.completions.create(
    model="x-ai/grok-build-0.1",
    messages=[{"role": "user", "content": "Write a FastAPI endpoint"}]
)

// Use your Vercel AI Gateway API key
// Model identifier: xai/grok-build-0.1
// Endpoint: https://gateway.ai.vercel.com/v1
import { generateText } from 'ai'
import { gateway } from '@ai-sdk/gateway'

const { text } = await generateText({
  model: gateway('xai/grok-build-0.1'),
  prompt: 'Review this code for security issues'
})

Pricing Comparison - Why $1/$2 Is Significant

At $1.00/$2.00, grok-build-0.1 is the cheapest coding-specialized model with always-on reasoning available via API today. Claude Haiku 4.5 matches on input at $1.00 but costs 2.5x more on output ($5.00 vs $2.00) and has no reasoning capability. Gemini 3.5 Flash has a larger context window (1M vs 256K) but costs 50% more on input and 4.5x more on output. For output-heavy workloads — long code generation, large file refactors, extended agent sessions — grok-build-0.1's $2.00 output rate is the most competitive number in this tier.

Native MCP Support - What It Means in Practice

Native MCP support means grok-build-0.1 can connect to any MCP server without a translation layer. Where many models require a middleware adapter to convert MCP tool calls into the model's native tool format, grok-build-0.1 understands the MCP protocol directly. For developers building agent pipelines on existing MCP infrastructure - using servers for GitHub, Linear, Notion, Google Workspace, file systems, or custom internal tools - the model plugs in without changes to the MCP server configuration.

xAI shipped a Connectors layer alongside Grok Build that includes pre-built integrations with GitHub, Notion, Linear, Google Workspace, Microsoft 365, Vercel, and Canva, plus Bring-Your-Own-MCP support for connecting any MCP-compatible tool. This integration catalog is narrower than Claude Code's 3,000+ MCP server ecosystem, but the pre-built connectors for the most common developer and professional tools cover most production agent workflows.

Decision Framework - When to Use grok-build-0.1 vs Claude Code vs Codex

Frequently Asked Questions