By xAI shipped Grok Build on May 14 — a terminal-based coding agent that enters a market already split between Claude Code and OpenAI’s Codex CLI. The timing isn’t accidental. Anthropic is changing Claude Code’s pricing on June 15. OpenAI just updated Codex to GPT-5.5. Developers are actively re-evaluating their AI coding stack. Grok Build wants to be the answer.
Here is what it actually brings, where it falls short, and which tool makes sense for what kind of work.
What Grok Build Is
Grok Build is a CLI tool — run it from your project folder, describe a task in plain language, and the agent gets to work. What separates it from the competition is the architecture underneath: Grok Build can spawn up to 8 concurrent sub-agents that work in parallel, each following a plan-search-build cycle. The model backing it, Grok 4.3 beta, runs a 16-agent Heavy architecture with a 2 million token context window — twice what Claude Code offers, five times what Codex CLI ships in its CLI form.
The other headlining feature is Plan Mode on by default. Before Grok Build touches a single file, it produces a full execution plan for the developer to review, comment on, or reject. Changes come back as clean diffs. Nothing is committed without approval. This is the single most-requested missing feature in Claude Code, and xAI ships it out of the box on day one.
The Benchmark Picture
Here is where Grok Build has a problem. SWE-bench Verified is the industry’s standard measure for whether an agent can actually resolve real GitHub issues. The current standings:
| Tool | Model | SWE-bench Verified | Context | Parallel Agents | Plan Mode Default |
|---|---|---|---|---|---|
| Codex CLI | GPT-5.5 | 88.7% | 400K (CLI) | No | No |
| Claude Code | Opus 4.7 | 87.6% | 1M tokens | No | No |
| Grok Build | Grok 4.3 beta | 70.8% | 2M tokens | 8 | Yes |
That is an 18-point gap on the metric that matters most. Nearly one in five tasks that Claude Code or Codex CLI resolve autonomously will fail with Grok Build. The 2M context and 8 parallel agents are real advantages — but they do not compensate for a benchmark gap of this size when you are routing actual production work through an agent.
It is worth noting: Grok Build is in early beta. Grok 4.3 is the first model xAI has purpose-built for agentic coding. This score will improve. Whether it improves enough, and how fast, is the open question.
Where Each Tool Actually Wins
Claude Code (Opus 4.7) is the maturity pick. It is stable in CI/CD pipelines, well-documented, and has the broadest ecosystem integrations. SWE-bench Pro jumped from 53.4% to 64.3% with Opus 4.7 — the most complex real-world issues are where it pulls ahead. If your team is running agents in production and reliability matters more than novelty, Claude Code is still the default choice. The June 15 pricing change is worth watching, but it does not change the fundamentals.
Codex CLI (GPT-5.5) wins on speed. At 240+ tokens per second — roughly 2.5x faster than Claude Code — it handles high-volume, boilerplate-heavy, or iterative editing tasks faster than either competitor. It also has a built-in review agent that critiques your diff before you commit, and the natively omnimodal backing model means you can feed it screenshots and mockups. For teams already on OpenAI subscriptions, there is no additional cost to start using it today.
Grok Build has two clear use cases: large mono-repos that actually need a 2M token context window, and investigation tasks that benefit from parallel exploration — debugging regressions, auditing an unfamiliar codebase, architecture analysis. The fan-out model genuinely shines when the right answer requires exploring multiple hypotheses at once. The plan-review workflow also makes it a reasonable fit for teams that want more control over what the agent does before it does it.
The $300/Month Question
Grok Build access requires a SuperGrok Heavy subscription: $99/month for the first six months, then $299–$300/month. Claude Code is bundled into Claude Pro and Team plans. Codex CLI is available on existing OpenAI subscriptions.
At $300/month, Grok Build is not a solo developer tool. It is a line item for a funded team treating AI-assisted development as a productivity investment. The $99 intro rate is aggressive customer acquisition designed to pull developers out of the Anthropic and OpenAI ecosystems — but the cliff at month seven is steep, and the benchmark gap means teams will need to validate the ROI before that clock runs out.
The Bigger Picture
The most important development here is not Grok Build itself — it is that three credible terminal-based coding agents are now competing head-to-head. Plan Mode, 2M context, and parallel sub-agents are features that Anthropic and OpenAI will now feel pressure to ship. The race to automate software engineering just added a third horse, and that is good for developers regardless of which tool they end up using.
For teams re-evaluating right now: Codex CLI is the speed pick, Claude Code is the reliability pick, and Grok Build is the one to watch once it reaches general availability and closes the benchmark gap. Check back in three months.
