xAI急推Grok Build衝到v0.2.11！挖角Cursor兩核心，企圖追上Claude Code

xAI Quickly Upgrades AI Coding Tool Grok Build from v0.1 to v0.2.11 in Less Than Half a Month, Adding 8 Proxy Parallelism, X Platform Search Integration, Plan Mode, and More, Aiming to Capture Market Share in AI Coding.
(Previous context: Building a $3 Trillion Super Giant? Rumors that Elon Musk Intends to Merge Tesla and SpaceX, Creating a New Entity Holding 30k BTC)
(Additional background: Altman Retracts "AI Will Destroy Humanity" Prediction: I'm Glad I Was Wrong, Is It Truly Sincere?)

Key Summary

• After launching on 5/14, xAI’s Grok Build was upgraded to v0.2.11 in half a month, supporting 8 proxy parallelism, X search integration, and Plan Mode enabled by default
• xAI recruited engineering and product leaders from Cursor, and obtained a $60 billion acquisition option for Cursor’s parent company, abandoning a $10 billion penalty
• SWE-bench Verified benchmark score is 70.8%, still significantly behind Claude Code at 87.6% and OpenAI Codex CLI at 85-88.7%

In just two weeks, xAI’s AI coding tool Grok Build, with 11 versions released in half a month, has completed most competitors’ half-year update cycles. This CLI-first agentic coding tool launched an early beta on 5/14, driven by their grok-code-fast-1 model, with a 256K token context window, supporting up to 8 sub-agents operating in separate Git worktrees concurrently.

It features Plan Mode enabled by default (requiring user approval before editing code), and is currently the only AI coding tool integrating real-time X platform search, allowing developers to directly query community discussions and track API disputes during development.

The latest update focuses on four areas: in agent capabilities, sub-agents can share terminal backends and task scheduling across conversations, with new “Laziness Detector” and proactive alert mechanisms; in compatibility, support added for Windows ARM64 and macOS x86_64, fixing copy-paste issues on Linux Wayland and WSL.

In terms of user experience, terminal video playback is now at 30FPS, supporting multi-image pasting and macOS screenshot shortcuts; for stability, default retry budget increased to about 5 minutes, strengthening timeout handling for background tools.

Poached Two Core Figures from Cursor

Relying solely on version numbers isn’t enough; xAI is also recruiting talent. They’ve brought in engineering leader Andrew Milich and product leader Jason Ginsberg from Anysphere, Cursor’s $50 billion valuation parent company, and also recruited Devendra Singh Chaplot from Mistral AI.

More notably, in April, xAI signed an option agreement with Anysphere, allowing them to acquire Cursor’s entire team and products later in 2026 for $60 billion; if they choose not to, they must pay a $10 billion penalty. Cursor is currently training models on xAI’s Colossus supercomputer in Tennessee.

Elon Musk admitted in March that “xAI didn’t get it right the first time, and we are rebuilding from the ground up,” with all 11 founding members of xAI having left.

SWE-bench Gap and AI Coding Goals

Grok Build scored 70.8% on SWE-bench Verified, while Anthropic’s Claude Code achieved 87.6%, and OpenAI Codex CLI reached 85-88.7%, a gap nearly a full generation. However, xAI’s strategy isn’t just about benchmark scores; supporting up to 8 sub-agents in parallel is the most among similar tools, and Grok Build can directly read Claude Code’s user skills, plugins, and hooks configuration files, enabling zero-cost migration.

A tester on Medium found that Grok Build automatically loaded 47 skills at startup, 12 of which came from Claude Code’s settings directory.

By 2026, the AI coding market is projected to reach $12.8 billion, with 84% of developers using or planning to use AI tools, averaging 3.1 tools per person. Cursor leads with $2 billion annual revenue and 1 million paying users; Claude Code holds 28% of the main tool market share and 46% of “favorite tool” votes; GitHub Copilot has 4.7 million paying subscribers.

Grok Build’s pricing follows xAI’s subscription model: SuperGrok at $30/month, X Premium+ at $40/month, and SuperGrok Heavy at $300/month (discounted to $99 for the first 6 months), unlocking full features. Arena Mode (automatic multi-agent competition scoring) is a flagship feature promoted, but as of v0.2.11, it’s not yet available.

Frequently Asked Questions

What is Grok Build? Is it paid?

xAI’s CLI-first AI coding tool, launched in early beta on 5/14. Requires subscription to SuperGrok ($30/month), X Premium+ ($40), or SuperGrok Heavy ($300) to use; no free plan available.

How does Grok Build compare to Claude Code?

Grok Build scored 70.8% on SWE-bench Verified, while Claude Code scored 87.6%, a difference of nearly 17 percentage points. However, Grok Build supports up to 8 sub-agents in parallel and can read Claude Code’s configuration files with zero migration cost.