Comparison
Grok Build vs Claude Code: which coding agent is better?
xAI built Grok Build in Claude Code's image, down to the mode names, and open-sourced it. Where it matches, and where the benchmarks say it doesn't.
Grok Build is a terminal coding agent from xAI, open source under Apache 2.0 since July 2026, and it's built in Claude Code's image: the same permission modes, a plan mode, subagents, skills, plugins, hooks and MCP. It's included in every Grok plan, the free one too, and runs Grok 4.7 by default or any model through custom backends. Claude Code needs a paid Claude plan, isn't open source and runs Claude Opus 5.5 by default. On independent tests Claude Code is ahead: Artificial Analysis scores it 68 to Grok Build's 56 on its coding agent index, and on the Terminal-Bench 4.0 leaderboard Claude Code with Fable 5.1 reached 57.9% to Grok Build with Grok 4.7's 37.6%. Grok Build is cheaper per task and faster.
Key takeaways
- Grok Build is open source, comes with every Grok plan including free (usage limits unpublished) and defaults to Grok 4.7.
- Claude Code needs Claude Pro or higher, from $17 a month, defaults to Opus 5.5 and isn't open source.
- Independent benchmarks favor Claude Code on hard terminal tasks; Grok Build ties on DeepSWE and costs about $9 a task to Claude Code's $14.
- Not the same Grok Build: xAI also uses the name for an app builder on grok.com.
| Grok Build | Claude Code | |
|---|---|---|
| Made by | xAI | Anthropic |
| License | Apache 2.0 | Proprietary |
| Default model | Grok 4.7 | Claude Opus 5.5 |
| Other models | Any provider or a local model | Claude models |
| Price | Every Grok plan, including free | Claude Pro from $17 a month |
| API price, default model (per million tokens) | $2 / $6 | $4 / $20 |
| Starts in | Ask mode | Auto mode (terminal, VS Code) |
| Coding Agent Index (Artificial Analysis) | 56 | 68 |
| Terminal-Bench 4.0 (leaderboard) | 37.6% with Grok 4.7 | 57.9% with Fable 5.1 |
| Cost per task (Artificial Analysis) | $8.82 | $14.19 |
What is Grok Build?
A terminal coding agent written in Rust, with a headless mode for scripts and ACP support for editors (xAI). It entered beta for SuperGrok and X Premium+ subscribers on May 25, 2026, was open-sourced on July 15, reached version 1.0 on August 7 and added Grok 4.7 on September 20 (changelog). It installs with one command on macOS, Linux or Windows. Its repository has about 27,000 GitHub stars and doesn't take outside contributions.
xAI also sells grok-build-0.1, a coding model on its API since May 29, 2026, and uses the name Grok Build for an app builder inside grok.com, open to every plan since August 19. This page compares the terminal agent.
How do their features compare?
- Permission modes: Grok Build's modes copy Claude Code's names: default (ask), acceptEdits, plan, auto, dontAsk and an always-approve mode. Claude Code starts in auto mode in the terminal and VS Code; Grok Build starts by asking.
- Plan mode: both. Grok Build's is read-only except for the plan file, which you approve, comment on or quit.
- Subagents, skills, plugins, hooks, MCP: both. Grok Build's subagents are on by default and can work in their own git worktrees.
- Sandbox: Grok Build's is off by default, with Landlock on Linux and Seatbelt on macOS when on.
- Where it runs: Grok Build is a terminal agent you can embed in editors. Claude Code also runs in VS Code, JetBrains, the Claude desktop app, the web and Slack.
- Models: Grok Build can use other providers' models through OpenAI- or Anthropic-style APIs, or a local model; Claude Code runs Claude models.
What do they cost?
Grok Build is included in every tier on xAI's pricing page: Free, SuperGrok Lite, SuperGrok at $30 a month, SuperGrok Plus at $100 with significantly higher usage, Heavy, Business and Enterprise, though xAI publishes no usage limits. With an API key you pay per token: Grok 4.7 costs $2 per million input tokens and $6 output, with a 500,000-token window, and grok-build-0.1 $1 and $2, with 256,000 (xAI).
Claude Code needs Claude Pro ($17 a month billed annually, $20 monthly), Max ($100 or $200) or a Team seat ($20 to $25). On the API, its default Opus 5.5 costs $4 and $20 per million tokens, and Sonnet 5.5 $2 and $10. Is Claude Code free?
Which is better on benchmarks?
Claude Code, on the hardest tasks. Artificial Analysis ran both itself, Claude Code with Sonnet 5.5 at max effort and Grok Build with Grok 4.7 at xhigh:
- Coding Agent Index: Claude Code 68, Grok Build 56.
- Terminal-Bench 4.0: 66% against 33%.
- DeepSWE: 72% against 73%.
- Cost per task: $14.19 against $8.82.
- Time per task: 1.5 hours against 39 minutes.
The Terminal-Bench 4.0 leaderboard agrees on the gap: Claude Code with Fable 5.1 at 57.9%, Grok Build with Grok 4.7 at 37.6% (17th), and Codex with GPT-6 Astra first at 58.2%. xAI publishes no benchmark for grok-build-0.1, and its own Grok 4.7 post shows Fable 5.1 ahead on Terminal-Bench 4.0 and CursorBench while Grok 4.7 edges it on DeepSWE.
Which should you use?
- The strongest results on long terminal tasks, and the most places to run it: Claude Code.
- An open-source agent you can read and change, or a free start: Grok Build.
- Any model, including a local one, with Claude Code's habits: Grok Build.
- Faster, cheaper tasks where top accuracy matters less: Grok Build.
More options are in Claude Code alternatives.
Can you use both?
In Poly (usepoly.co), a room can mix labs: Claude models run through Claude Code, and Grok models, including the free grok-build-0.1 and the paid Grok 4.7, run through Codex, all in one workspace that everyone in the room shares. Each person picks the model for their own turns. Multi-model AI coding.
Common questions
Is Grok Build better than Claude Code?
Not on independent tests of hard tasks: Artificial Analysis scores Claude Code 68 to Grok Build's 56, and Terminal-Bench 4.0 shows a similar gap. Grok Build is cheaper and faster per task and ties on DeepSWE.
Is Grok Build free?
It's included in every Grok plan, the free one too, though xAI doesn't publish usage limits, and the agent itself is open source. With an API key you pay per token.
Is Grok Build open source?
Yes, under Apache 2.0 since July 15, 2026. xAI doesn't accept outside contributions, and binaries come from its installer.
What model does Grok Build use?
Grok 4.7 by default, according to xAI. It can also use grok-build-0.1, other providers' models through OpenAI- or Anthropic-style APIs, or a local model.