Grok Build — xAI ships a terminal coding agent on the hosted grok-build-0.1 model

xAI opened Grok Build to public beta via the xAI API in early June — a Rust, terminal-native coding agent and CLI (xAI's answer to Claude Code, Codex, and Gemini CLI), powered by the new grok-build-0.1 model. It is a fast agentic coding model: 256K context, text + image in, function calling + structured outputs + reasoning, at $1 / 1M input and $2 / 1M output ($0.20 cached). The catch for a local-AI site: grok-build-0.1 is hosted / API-only — no open weights — so it shifts the cloud landscape, not your local picks.

Verdict: xAI's terminal coding agent + a cheap hosted coding model — API-only, so no local pick changes

The take

The facts, verified against xAI's own docs and pricing: grok-build-0.1 was created on May 20 (early access), the Grok Build CLI was introduced on May 25, and the model reached public beta via the xAI API in early June. It is hosted only — available on the xAI API, OpenRouter, and third-party routers (Requesty, Kilo Code) — with no Hugging Face weights. xAI has only ever open-sourced Grok-1, back in 2024; grok-build-0.1 is not open. Specs: 256K context, text + image input / text output, $1.00 input / $2.00 output / $0.20 cached per 1M (requests over 200K context are billed higher). xAI has not published benchmarks for this variant yet, so treat capability claims as unproven for now.

What it actually is: a terminal-native coding agent — the same shape as Claude Code, OpenAI Codex, Gemini CLI, and Microsoft's MAI-Code-in-Copilot — running on a purpose-built coding model. xAI also just listed a new flagship, grok-4.3 (1M context, $1.25 / $2.50), on its API. So within a two-week window, Anthropic, OpenAI, Google, Microsoft (MAI), and now xAI have all shipped the same pairing: a cheap fast first-party coding model plus an agentic CLI. The whole frontier is converging on that pattern.

The local-vs-cloud signal for our readers is price. At $1 / $2 per 1M (≈$1.50 blended at a 50/50 split), grok-build-0.1 sits in the same cheap-hosted-coding bracket as MAI-Code-1-Flash and Gemini 3.1 Flash-Lite. That is the trend worth tracking: the cloud floor for "good enough" agentic coding keeps dropping, which raises the bar a local coding rig has to clear on the things cloud cannot give you — privacy, offline, and zero marginal cost at high volume — rather than on raw capability. With no published benchmarks, we cannot yet say where grok-build-0.1 lands against the local coding picks.

Our call: no models.js or planner change — there is nothing to run locally. We are also holding the cost calculator for now and will add the cluster of cheap hosted coding models together (grok-build-0.1, MAI-Code-1-Flash, grok-4.3) in one dedicated pass once they are benchmarked, rather than piecemeal — the pricing here is clean and final ($1 / $2), but they belong in the chart as a set. The best local coding picks (Qwen3-Coder-30B-A3B, Qwen 3.6-35B-A3B, GLM-5.1) are exactly where they were.

Where this fits

Models: Qwen3-Coder-30B-A3B · Qwen 3.6-35B-A3B · GLM-5.1 · DeepSeek V4-Flash

Hardware: NVIDIA RTX 4090 · RTX 5060 Ti 16 GB · Dual RTX 5090

Sources

Next step

Try this in the planner→