Elon Musk posted on the social network X a request for critical feedback on Grok Build. In doing so he was responding to an enthusiastic post by one of the users. The reaction turned out to be not so optimistic. Dozens of comments were written under the billionaire's posts. A significant share of the authors pointed to the product's weak spots.

We explain what is wrong with Grok Build and why Anthropic, in the opinion of Musk's subscribers, is ahead.

Want more exclusive news and analysis? Subscribe to our Telegram channel, discuss the news and share your thoughts on the market's future in the chat!

What Grok Build is

Grok Build is xAI's agentic CLI tool for development. It was launched in early beta in May 2026. The utility works directly from the terminal and is available only to SuperGrok and X Premium Plus subscribers. The cost of access is about $300 per month, which puts the product in the same price range as Claude Code and GitHub Copilot.

For complex tasks, the developers provided a planning mode. The user can approve a plan, comment on individual steps, or rewrite it entirely before execution begins. After that, every change is shown as a diff (a line-by-line display of the differences between two versions of a file).

The base model, Grok 4.3 beta, uses an architecture with 16 agents and a 2 million token context window. The tool is capable of running up to eight parallel agents.

The comparison with competitors does not favor xAI

The most frequent topic in the comments was the comparison with Claude Code and OpenAI Codex. Unfortunately, these comparisons almost always turned out to be not in Grok's favor. For example, one developer described a direct experiment. Grok worked on implementing a project for almost two days, after which the same volume of tasks was switched to Codex. As a result, the competitor advanced twice as far in six hours.

Another user claimed that Grok went into endless loops for thirty minutes. At the same time, Opus solved the same problem on the first attempt. A third specialist noted that the inference speed in Grok CLI feels weak and slow compared with Claude Code and Codex. Because of this, watching the agent work becomes uncomfortable.

Source: BeInCrypto