xAI just put a coding agent into developers’ hands. The company released grok-build-0.1 on its API in public beta, a model built specifically for agentic software engineering, according to xAI’s announcement. This isn’t a general chat model that answers coding questions on the side. It’s trained to act as an autonomous engineer that can plan, write, refactor, and keep iterating until a task is done.
What stands out here is the positioning. xAI is going straight at the coding-agent market that Anthropic’s Claude Code and others have been defining, and it’s pricing aggressively to do it.
What Grok Build 0.1 actually does
Here’s what xAI and early documentation describe:
- Autonomous engineering, not Q&A. The model plans multi-step work, invokes tools, generates structured output, and runs in continuous loops until the job finishes. Think refactoring a large codebase rather than answering a single snippet question.
- Big context window. It handles a 256K-token context with text and image input, which is enough to hold sizable chunks of a real project in working memory.
- Fast output. xAI reports the model serves at 100-plus tokens per second. For an agent running long loops, speed compounds, since every step waits on the one before it.
- Native tool use and reasoning. Tool invocation and reasoning chains are built in, not bolted on, which matters for agents that need to call build systems, run tests, and react to results.
- A CLI to drive it. xAI followed the API release with a command-line installer, including a Windows PowerShell version, so developers can run the agent directly in their terminal.
How it compares
The obvious reference points are Anthropic’s Claude Code and the wave of terminal-based coding agents that shipped over the past year. Grok Build 0.1 plays in the same arena: a model tuned for autonomous engineering, paired with a CLI, aimed at developers who want an agent loop rather than a chatbot.
The sharper angle is price. xAI lists API rates at $1.00 per million input tokens and $2.00 per million output tokens, with cached input at $0.20 per million. That undercuts a lot of the frontier coding models developers currently reach for. If the quality holds up in real projects, cost alone will get teams to test it.
Availability and access
A few things to know before you plan around it:
- It’s in public beta on the xAI API, so expect changes and rough edges.
- Early access is available to SuperGrok and X Premium+ subscribers.
- Open-source agents can call grok-build-0.1 through OAuth using existing Grok subscription credentials, which lowers the friction for anyone already wired into that ecosystem.
The caveats
The version number is doing real work here. “0.1” signals an early release, and xAI is calling it a beta. Specs and pricing can shift, and a model trained to run autonomous loops needs careful sandboxing before you point it at production code. There’s also a gap worth flagging: some third-party writeups cite larger context figures than the 256K xAI documents, so treat any number that doesn’t come straight from xAI with caution until it’s confirmed.
Why it matters
This is significant because it puts xAI on the board in the coding-agent race, not just the chatbot race. The combination of agentic training, a terminal CLI, fast token throughput, and low API pricing is a direct play for developer workflows. The companies that own the agent layer own where a lot of real engineering work will happen next.
Whether grok-build-0.1 earns a spot in daily workflows comes down to reliability on messy, real-world code, the part no spec sheet can promise. The beta tag means xAI knows that too. For developers who want to kick the tires, full details and pricing are available at the original source.
Sources: Grok Build 0.1 on API | xAI, xAI Docs, OpenRouter