Grok Code and Grok Build are not two competing products — they are the same model under two names. xAI's own documentation lists grok-code-fast-1 and grok-code-fast as aliases of grok-build-0.1, which means an API call to either identifier reaches the same weights. If you have been trying to work out which one to pick, that decision does not exist. The decision that does exist is a different one: Grok Build 0.1, the agentic coding model, versus Grok 4.5, xAI's general-purpose flagship.
TL;DR — Key Takeaways
- Same model:
grok-code-fast-1is a documented alias ofgrok-build-0.1. "Grok Code" is the earlier name that stuck around. - The real choice: Grok Build 0.1 (agentic, works inside your repo) vs Grok 4.5 (general chat, answers coding questions).
- For agentic coding: Grok Build 0.1 — 256K context, always-on reasoning, function calling, structured outputs, $1/$2 per million tokens with $0.20 cached input.
- For asking questions: Grok 4.5, xAI's July 2026 frontier flagship, which is what you get by default on grok.com.
- Watch the naming: older tutorials and integrations still say "grok-code-fast-1". They are not out of date in a way that breaks — the alias still resolves.
Why the confusion exists
xAI renamed its coding model mid-flight. Before Grok Build shipped in May 2026, the coding-specific model was grok-code-fast-1 — a name that appeared in API examples, third-party integrations, OpenRouter listings and a lot of tutorials. When xAI launched the Grok Build CLI, the model became grok-build-0.1, and the older identifiers were kept as aliases so nothing broke.
That is good engineering and bad naming. The old name never disappeared from the ecosystem, so people reasonably assume there are two products and go looking for a comparison. There is nothing to compare. What people actually need answered is which xAI model should I point at my code — so that is what the rest of this page does.
The three names you will run into
| Grok Code grok-code-fast-1 |
Grok Build 0.1 grok-build-0.1 |
Grok 4.5 | |
|---|---|---|---|
| What it is | Alias of the model on its right | Agentic coding model | General frontier flagship |
| Powers | — | Grok Build CLI | grok.com, X, SuperGrok |
| Context window | 256K (same model) | 256,000 tokens | Not comparable — different tier |
| API price (in / out) | Same as Grok Build | $1 / $2 | See xAI pricing |
| Cached input | Same as Grok Build | $0.20 / 1M | — |
| Edits files in a repo | Yes (same model) | Yes | No — chat only |
| Use it for | Legacy integrations | Agentic coding | Coding questions, general work |
Grok Build 0.1 specs and pricing are from xAI's model documentation. Grok 4.5 is xAI's general-purpose flagship launched in July 2026; xAI positions it as Opus-class on coding and reasoning at lower cost, which is a vendor claim rather than an independently verified benchmark. For current token maths on any of these, use the Token Cost Calculator.
Can Grok code?
Yes — but "Grok" is a family, not a single model, and the family members are aimed at different jobs. Grok 4.5 writes code in a chat window: you describe a problem, it returns a snippet, you paste it. That is the same shape as asking ChatGPT or Claude a coding question, and for a quick function or a debugging second opinion it is perfectly good.
Grok Build 0.1 does something categorically different. It was trained for agentic software-engineering workflows rather than general chat: it reads your codebase, plans a change, edits files, runs tools, and iterates through the loop until the task is done. Reasoning is always on. It supports function calling and structured outputs, which are the table stakes for driving an agent loop at all. If what you want is an AI that touches your repository rather than one that describes what you should type, that is the split that matters.
Which Grok model is best for coding?
Work backwards from where the code lives.
Use Grok Build 0.1 (via the Grok Build CLI) if you:
- Want the model working inside a real repository — reading files, making multi-file edits, running your tests.
- Run high-volume agent loops where $1/$2 per million tokens, and especially the $0.20 cached-input rate, changes your monthly bill materially.
- Already pay for SuperGrok or X Premium Plus, which bundle the CLI at no extra cost.
- Have an existing Claude Code setup — Grok Build is deliberately compatible with
CLAUDE.mdfiles, skills, MCP servers, plugins and hooks, so trying it costs you very little.
Use Grok 4.5 if you:
- Mostly ask coding questions rather than delegate coding work.
- Want one model for code plus everything else — research, writing, reasoning about live posts on X.
- Are not ready to give an agent write access to a repo, which is a completely reasonable position.
The question is not "Grok Code or Grok Build". It is "do I want an agent in my repo, or an expert in my chat window?" Those are different products, and only one of them is Grok Build.
Is Grok Build 0.1 good?
Good at what it is aiming for, not at topping the leaderboard. The most-cited figure for the model is ~70.8% on SWE-Bench Verified, and two caveats travel with it: it comes from xAI's own internal harness rather than an independent run, and it sits roughly 15–18 points below the frontier agents on the same benchmark. For a release explicitly numbered "0.1" that is a respectable showing, and it tells you exactly where the model competes — on speed, token price and ecosystem fit, not on raw capability.
We go through the benchmarks, the CLI's plan mode, the parallel sub-agents and the honest cost caveats in the full Grok Build 0.1 review, which is the canonical page for everything about the agent itself.
The bottom line
There is no Grok Code versus Grok Build decision to make, because grok-code-fast-1 and grok-build-0.1 are the same model. If an older tutorial or integration hands you the legacy identifier, it will still resolve — nothing is broken. The decision worth thinking about is whether you want xAI's agentic coding model (Grok Build 0.1, in your terminal, priced for volume) or its general flagship (Grok 4.5, in a chat window, better at everything that is not repository work). Most developers who have tried both end up keeping a frontier agent for the hardest jobs and reaching for Grok Build on the cost-sensitive ones.
Related: the full Grok Build 0.1 review, Grok Build 0.1 vs GPT-5.5, Grok Build 0.1 vs Claude Code, and Cursor vs GitHub Copilot. Still deciding? Run Pickurai's free finder.
Common questions
Is Grok Code the same as Grok Build?
Effectively yes. Per xAI's documentation, grok-build-0.1 carries the aliases grok-code-fast-1 and grok-code-fast. They are not two competing products — "Grok Code" is the earlier name for the coding model that now ships as Grok Build 0.1, and an API call to either identifier reaches the same model.
Which Grok model is best for coding?
For agentic coding — reading a repository, editing files, running tools in a loop — use Grok Build 0.1 through the Grok Build CLI. It is the model xAI trained for software-engineering workflows, with a 256,000-token context window, always-on reasoning, function calling and structured outputs. For asking coding questions in a chat window, xAI's general flagship Grok 4.5 is the one you are already talking to on grok.com.
Can Grok write code?
Yes, but the models are aimed at different jobs. Grok 4.5 answers coding questions conversationally. Grok Build 0.1 is the agentic model built to work inside a repository — it plans a change, edits files, runs tools and iterates. If you want an AI that actually touches your codebase rather than pasting snippets into a chat, Grok Build 0.1 is the one.
How much does Grok Build 0.1 cost?
On the xAI API it is $1.00 per million input tokens, $2.00 per million output tokens, and $0.20 per million cached input tokens. The Grok Build CLI is bundled at no extra cost with SuperGrok and X Premium Plus. Your all-in cost depends on your usage pattern and how well prompt caching works on your repository, so model it with the Token Cost Calculator before switching.
Why do people search for "Grok Code" if it isn't a separate product?
Because it was one first. grok-code-fast-1 was xAI's earlier coding model, and the identifier survives as an alias plus in older tutorials, API examples and third-party integrations. The name outlived the standalone product, which is why "grok code vs grok build" and "which grok model is best for coding" keep showing up as searches.