Affiliate Disclosure: Some links on this page are referral links. If you sign up through them, we may earn a commission at no extra cost to you. Our findings are not influenced by commissions.
Research note: This is a research-based review. We did not run a hands-on Grok Build test for this article — every benchmark figure below is attributed to its publisher (xAI, Artificial Analysis, or independent testers), and where sources disagree we say so. Pricing and plan details were verified against sources checked October–November 2026; xAI changes these numbers, so re-check x.ai before you pay.
Verdict Up Top
Grok Build is xAI’s terminal-native coding agent: a fullscreen CLI that reads your repo, plans a change, and edits files — with up to eight subagents racing the same problem in parallel. It launched in early beta in May 2026 behind a $300/month paywall, went open source in July 2026, and by October 2026 was free to try, bundled into the X Subscription from $8/month.
The short version: Grok Build is the cheapest way into a frontier-class terminal coding agent, and its cost-per-task numbers are genuinely impressive — independent testing puts it at $2.49 per resolved task, against $5.07 for Codex and $11.80 for Claude Code. But it trails the leaders on raw capability: the headline 70.8% SWE-Bench Verified figure floating around launch coverage actually belongs to its predecessor model, and on harder benchmarks (DeepSWE, SWE-Bench Pro) the gap to Claude Code and Codex is real. Add a July 2026 incident in which one version silently uploaded entire git repos — history included — to xAI’s cloud, and the verdict is: a strong budget pick for scoped, well-defined coding work, not yet the tool to bet a complex codebase on.
Who should try it: developers already paying for an X subscription, anyone who wants a cheap API-priced coding agent, and teams that value the open-source CLI. Who should skip it for now: anyone handling sensitive proprietary code without first checking the current data-retention terms, and anyone whose daily work is large multi-file refactors — the benchmark gap shows up exactly there.
What Is Grok Build?
Grok Build is xAI’s agentic coding CLI. You point it at a project directory, describe a task in plain English, and it inspects the repository, locates the relevant files, and proposes and applies changes — all inside a fullscreen terminal UI with mouse support.
Two products now share the name, and most confusion about Grok Build comes from mixing them up:
- Grok Build (terminal CLI) — the original product, launched in early beta on May 14, 2026 (formal launch post May 25, 2026). This is the coding agent this review covers.
- Grok Build (in the Grok app) — an app builder inside Grok chat on web, iPhone, and Android. Introduced July 28, 2026 as “Build Mode” for SuperGrok Heavy subscribers; renamed Grok Build and opened to every plan on August 19, 2026.
The CLI is written in Rust and has been open source under Apache 2.0 since July 15, 2026, at github.com/xai-org/grok-build, where it gathered roughly 27,000 stars in its first months. One caveat: xAI does not accept external pull requests, so it is open source in the license sense, not community-governed.
Under the hood, the CLI launched with a purpose-built coding model, grok-build-0.1 (256K-token context window, released on the xAI API May 29, 2026). xAI has since moved the CLI to run on its flagship models — Grok 4.6 by default, with Grok 4.7 powering it according to sources checked in October 2026.
The Parallelism Bet: Up to 8 Subagents
Grok Build’s headline architectural idea is parallelism over depth. Where Claude Code runs one powerful reasoning pass, Grok Build can spawn up to eight subagents that run simultaneously on the same task — planning, searching documentation, and writing code in parallel, then reconciling the results.
In theory this trades single-agent reasoning quality for speed and breadth: eight attempts at a problem surface more approaches, and the best one wins. In practice, whether that beats one stronger agent depends on the task. For well-scoped work — a bug fix, a small feature, a test-writing pass — parallel attempts are a reasonable strategy. For the kind of deep, multi-file refactor where a wrong first step poisons everything downstream, eight parallel wrong first steps are not obviously better than one careful one.
The CLI also ships the standard agentic-coding toolkit: a Plan Mode that blocks edits until you approve the plan, MCP server support, AGENTS.md conventions, hooks, skills, sandboxed execution, a headless mode for scripts and CI/CD pipelines, and Agent Client Protocol (ACP) integration for other apps.
Grok Build Pricing: From $300/Month to Free to Try
No coding agent has repriced as fast as Grok Build. The timeline:
| Date (2026) | What changed |
|---|---|
| May 14 | Early beta launches, SuperGrok Heavy only — $300/month, with a six-month introductory price of $99/month |
| July 15 | CLI open-sourced (Apache 2.0) |
| Aug 19 | In-app Grok Build reaches every plan |
| Sep 21 | Grok 4.7 becomes the model behind Grok Build |
| Oct 12 | The X Subscription launches for new buyers; the Free tier gets a limited Grok Build allowance |
Current access (as of November 2026): the official Grok Build page says it is “available to try for free.” More usage comes through the X Subscription, where Grok Build shares one usage pool with Grok chat, voice, Imagine, and API access: Plus $8/month, Premium $30/month, Super $100/month, Ultra $200/month, with the usage pool scaling by tier (Premium roughly 4× Plus, Super 15×, Ultra 50×). Annual billing is cheaper — Plus $80/year, Premium $300/year, Super $1,000/year, Ultra $2,000/year.
That makes SuperGrok at $30/month the lowest entry price in the terminal-agent category, and the free tier means you can evaluate it without paying anything — a complete reversal from the $300 launch price. If you already pay for an X subscription, Grok Build is effectively bundled.
API pricing: Grok Build also runs on an xAI API key, billed per token. The grok-build-0.1 API model is listed at $1.00 per million input tokens ($0.20 cached) and $2.00 per million output tokens, with prompts over 200K tokens costing double (per GrokMag’s check, October 18, 2026). Note that launch-era coverage reported different figures ($0.20 input / $1.50 output) — the API price has moved, so verify the current rate on x.ai before budgeting a heavy workload. The flagship Grok 4.7 model, which now powers the CLI, is listed at $2 per million input tokens and $6 per million output tokens.
For the broader Grok pricing picture, see our Does Grok Cost Money? SuperGrok Explained guide.
Benchmarks: What the Numbers Actually Say
This is where honest reporting matters, because the most-quoted Grok Build number is misleading.
The 70.8% SWE-Bench Verified figure appears in launch coverage as Grok Build’s score. According to GrokMag’s October 2026 check, that score actually belongs to grok-code-fast-1, Grok Build’s predecessor model, measured on xAI’s own harness in August 2025. xAI has published no SWE-Bench Verified score for grok-build-0.1 itself. Treat any article that presents 70.8% as grok-build-0.1’s score as wrong.
Independent figures for grok-build-0.1 (as reported by Agentriot, June 2026): 71.4% on Vals AI SWE-bench, 50.6% on Terminal-Bench 2.0, and 88.9% on PinchBench. Respectable, not leading.
The more useful comparison is Artificial Analysis’s Coding Agent Index, which pairs specific models with specific coding harnesses and averages results across DeepSWE, Terminal-Bench 2.1, and SWE-Atlas-QnA. On that index, Grok 4.5 running inside Grok Build scores 76 points — one point behind the leading setup and roughly matching GPT-5.5 in Codex. The per-task breakdown is more revealing:
| Benchmark | Grok 4.5 in Grok Build | Best rival |
|---|---|---|
| Terminal-Bench 2.1 | 83.3% | ~84.3% (leading model) |
| DeepSWE 1.1 | 53% | 70% (leading model) |
| SWE-Bench Pro | 64.7% | 80.4% (leading model) |
| AutomationBench-AA | 51.4% | 48.6% (next best) — Grok 4.5 leads here |
The pattern: Grok Build is competitive on terminal-style and automation tasks, and clearly behind on the hardest reasoning-heavy coding benchmarks. DeepSWE and SWE-Bench Pro are the tests that most resemble real multi-file engineering work, and that is exactly where the gap opens.
The cost side flips the story. Artificial Analysis’s cost-per-resolved-task figures: Grok Build $2.49, Codex $5.07, Claude Code $11.80. Grok 4.5 averages roughly 1.9 million tokens per task, against 6.2 million for GPT-5.5 and 7.2 million for the leading Anthropic model — it uses far fewer tokens to get a slightly worse result, which is why it wins so decisively on price. On AutomationBench-AA it also leads outright (51.4%) at $0.34 per task, against $1.35 and $1.46 for its rivals.
Bottom line on benchmarks: if your work is scoped tasks and automation, Grok Build delivers near-frontier results at roughly a third of the cost. If your work is hard, multi-file engineering, the benchmark gap is real and will show up as more failed attempts and more human review time.
The July 2026 Repo-Upload Incident
Any honest Grok Build review has to cover this. In mid-July 2026, an independent researcher found that version 0.2.93 of the CLI uploaded entire repositories — Git history included — to xAI’s cloud storage when run in a directory, at roughly 27,800× the data a task needed. The upload included files and secrets the task never touched.
xAI switched the behavior off on July 13, 2026, said default data retention has been off for all Grok Build users since July 12, and said retained coding data is being deleted. No independent audit has publicly confirmed the deletion.
What this means for you:
- Do not run any coding agent — Grok Build included — against a repo containing secrets (API keys, credentials, private keys) unless you have confirmed the current data-handling terms. Use sandboxed execution and a clean checkout.
- Read the current privacy terms before first use. Grok Build on the free tier authenticates through grok.com, which is the consumer product covered by xAI’s consumer privacy policy — not the enterprise Data Processing Addendum. Enterprise SSO and enterprise deployment documentation now exist, but the free tier’s coverage is the consumer policy.
- The incident is fixed, not forgotten. It is the single biggest reason to be careful with Grok Build on proprietary code, and the reason we flag it before the feature list rather than after it.
Our broader privacy guide for Grok covers the same ground: Is Grok AI Safe? Privacy & Data (2026).
Grok Build vs Claude Code vs Codex
| Grok Build | Claude Code | Codex CLI | |
|---|---|---|---|
| SWE-Bench Verified (reported) | 70.8%* | ~87.6% | ~85–88.7% |
| Approach | Terminal-native, up to 8 parallel subagents | Single powerful agent | Cloud-sandboxed parallel agents |
| Cheapest entry | Free to try; $8/mo via X Plus | $20/mo | $20/mo |
| API cost per resolved task | $2.49 | $11.80 | $5.07 |
| Open source | Yes (Apache 2.0, no external PRs) | No | Partially |
| Best for | Scoped tasks, automation, budget | Complex multi-file work | Balanced, sandboxed runs |
* The 70.8% figure belongs to predecessor model grok-code-fast-1 on xAI’s own harness, not to grok-build-0.1 — see the benchmarks section above.
Claude Code remains the capability leader for hard engineering work, and its benchmark lead is large enough to matter. Codex sits in the middle. Grok Build’s case is price and openness: it is the cheapest frontier-class terminal agent, the only one of the three that is open source, and the only one you can try for free without a card.
For a broader coding-agent comparison, see our Best Uncensored AI for Coding Help guide.
How to Install Grok Build
The install is a one-liner from xAI’s CLI page:
curl -fsSL https://x.ai/cli/install.sh | bash
First launch opens a browser to authenticate via grok.com (enterprise SSO is supported). After that, run grok in a project directory and describe the task in plain English. Use Plan Mode before letting it edit anything on a repo you care about, and run it in a clean checkout with no secrets present — see the incident section above.
If you decide to cancel later, our How to Cancel SuperGrok (2026) guide covers the subscription side.
Who Should Use Grok Build?
Try it if you:
- Already pay for an X subscription — it is bundled, so the marginal cost is zero.
- Do mostly scoped coding tasks: bug fixes, small features, test-writing, automation scripts.
- Want the cheapest per-task cost among the major terminal agents.
- Value an open-source CLI you can inspect and run against your own inference.
- Want to evaluate a coding agent without paying first — the free tier exists.
Skip it (for now) if you:
- Work daily on large, complex multi-file refactors — Claude Code’s benchmark lead is worth the $20.
- Handle sensitive proprietary code and cannot verify the current data-retention terms.
- Need mature enterprise governance, documentation, and support — Grok Build is still an early beta, and its docs are thinner than its rivals’.
The Bottom Line
Grok Build is a real coding agent with a real price advantage and a real benchmark gap. Six months after launch it went from a $300/month gated beta to a free-to-try tool bundled into an $8/month subscription — the fastest repricing in the category, and a sign xAI knows the $300 price was the problem. The parallel-subagent architecture is a genuine idea, the open-source CLI is a genuine differentiator, and the $2.49-per-resolved-task cost is the best in its class. But on the hardest coding benchmarks it trails Claude Code and Codex by a wide margin, and the July repo-upload incident means it has trust to rebuild with anyone running it on proprietary code.
Our verdict: worth trying free, worth paying $8–$30/month for if your work is scoped tasks and automation, not yet a Claude Code replacement for complex engineering.
Try Grok Build free → — or compare it with the chatbot side of Grok in our Grok vs ChatGPT: The Honest 2026 Comparison.
FAQ
Is Grok Build free?
Yes, to try. The official Grok Build page says it is “available to try for free,” and since October 12, 2026 the Free tier of the X Subscription includes a limited Grok Build allowance. More usage comes with Plus ($8/month), Premium ($30/month), Super ($100/month), or Ultra ($200/month). API access is billed per token instead.
What model powers Grok Build?
It launched with a dedicated coding model, grok-build-0.1 (256K context). xAI has since moved the CLI to its flagship models — Grok 4.6 by default, with Grok 4.7 powering it according to sources checked in October 2026.
How good is Grok Build at coding?
Competitive on scoped and automation tasks, behind on hard multi-file engineering. Independent figures: 71.4% on Vals AI SWE-bench and 50.6% on Terminal-Bench 2.0 for grok-build-0.1; Grok 4.5 in Grok Build scores 76 on Artificial Analysis’s Coding Agent Index, with 53% on DeepSWE and 64.7% on SWE-Bench Pro — both well behind the leaders. Its strength is cost: $2.49 per resolved task versus $5.07 for Codex and $11.80 for Claude Code.
Is Grok Build open source?
Yes. The CLI source has been published under Apache 2.0 since July 15, 2026 at github.com/xai-org/grok-build (roughly 27,000 stars). xAI does not accept external pull requests, so it is open source by license rather than community-governed.
Is Grok Build safe to run on my code?
It is safer than it was in July 2026, when version 0.2.93 uploaded entire repos — Git history included — to xAI’s cloud. xAI says that behavior was switched off July 13, 2026 and default retention has been off since July 12. Even so, do not run it against a repo containing secrets without checking the current terms, and prefer sandboxed execution and a clean checkout.
Grok Build vs Claude Code — which should I choose?
Claude Code if your daily work is complex multi-file engineering — its benchmark lead (~87.6% SWE-Bench Verified) is large. Grok Build if your work is scoped tasks and automation, you already pay for X, or you want the lowest cost per task ($2.49 vs $11.80) and an open-source CLI.
How do I install Grok Build?
curl -fsSL https://x.ai/cli/install.sh | bash, then authenticate via grok.com on first launch. Use Plan Mode before letting it edit files on a repo you care about.
What is the difference between Grok Build and Build Mode?
They are the same product line at different stages. “Build Mode” was the July 2026 name for the app builder inside Grok chat; on August 19, 2026 it was renamed Grok Build and opened to every plan. The terminal coding agent keeps the same name, which is why the two get confused.
Sources consulted for this review (November 2026): xAI Grok Build launch post and CLI page (x.ai/cli); GrokMag, “Grok Build: Coding Agent Features, Plans and Price” (checked October 18, 2026); Developers Digest, “Grok Build Developer Guide” (re-verified October 22, 2026); Fello AI, “Grok Build: What It Is, Pricing and How to Install It”; Artificial Analysis Coding Agent Index and Grok 4.5 benchmark reporting; Agentriot independent benchmarks for grok-build-0.1 (June 2026); VibeCompare Codex vs Grok Build pricing (verified November 4, 2026); AIToolGrade Grok Build review; The Weighted Average launch coverage. Benchmark figures are as reported by their publishers; xAI has published no SWE-Bench Verified score for grok-build-0.1 itself. Prices and plan details change — verify the current terms at x.ai before subscribing.