Theo Calls Out Grok 4.7 for Losing Speed and Efficiency

Headline

Theo criticizes Grok 4.7 for trading away speed, cost efficiency, and usability without delivering a clear capability gain.

Summary

In a widely circulated critique, Theo of T3.gg says Grok 4.7 is 30–80% less token-efficient than claimed, slower, and more than twice as expensive in real-world use than Grok 4.6. He also reports weaker performance on some benchmarks, poor frontend and nonexistent 3D capabilities, and recurring output loops, while acknowledging that the model remains useful for certain engineering tasks.

Analysis

Theo points out that Grok 4.7 burns more tokens, runs slower, and costs over twice as much as the last version for everyday work. Some benchmarks actually dropped. The model struggles with frontend code and has no 3D support at all, and it sometimes repeats itself in loops. It still handles certain engineering tasks without issues.

This shows how a model can look better on paper yet feel worse in practice because of speed, token use, and reliability problems. If the cost numbers hold, Grok 4.7 could lose ground to cheaper options like Astra on large coding or agent projects. The frontend and 3D complaints point to gaps when AI moves past plain text into building software or interactive tools. This comes from one experienced user rather than a full test, so it mainly flags frustration among heavy users rather than proving a broad drop in quality. It does show the risk of focusing on benchmark scores while letting day-to-day usability slip.

Impact Assessment

• Significance: Medium
• Categories: Technical Insight, Industry Trend, Market Impact