Grok Code Fast

Grok Code Fast 1 — Ultra-Fast AI Model for Agentic Coding

Grok Code Fast 1 is a speedy and economical reasoning model that excels at agentic coding. The model was built from scratch, starting with a brand-new architecture.

The model uses a 314B parameter Mixture-of-Experts (MoE) architecture. Grok Code Fast 1 features a context window of 256K tokens. It's probably the fastest model out there right now.

Key Features:

• Exceptionally versatile across the full software development stack, particularly adept at TypeScript, Python, Java, Rust, C++, and Go
• Can complete common programming tasks with minimal oversight, ranging from building zero-to-one projects to performing surgical bug fixes
• Transparent reasoning process — the model provides a clear trace of its thought process, showing which tools it used and why
• Prompt caching optimizations, regularly achieving cache hit rates above 90%

• On the full subset of SWE-Bench-Verified, scored 70.8%

  
  One of the standout features of Grok Code Fast 1 is its focus on the interactive development loop. Instead of providing a single, perfect solution to a complex problem, it excels at rapid, iterative tasks. The model includes optimization for tool usage commands like grep, terminal operations, and file editing, and is meant to integrate smoothly with coding environments and IDEs.