xAI unveils grok-code-fast-1
xAI has released grok-code-fast-1, a new reasoning model built specifically for agentic coding.
Created on August 28|Last edited on August 28
Comment
xAI has released grok-code-fast-1, a new reasoning model built specifically for agentic coding. Many large language models can write code, but developers often find them slow when handling iterative workflows that require multiple loops of reasoning and tool use. grok-code-fast-1 was designed to address this need by offering a faster, more responsive model. It was trained on a corpus rich in programming content and refined on curated datasets of real-world pull requests and coding tasks.
Integration with coding platforms
The model was built with launch partners to ensure seamless performance inside common developer tools. It has mastered standard utilities such as grep, terminals, and file editing, making it feel at home in modern IDEs. grok-code-fast-1 is being offered for free for a limited time on GitHub Copilot, Cursor, Cline, Roo Code, Kilo Code, opencode, and Windsurf.
Performance and speed
Speed is the defining feature of this release. xAI’s supercomputing team developed new serving techniques that let the model process tool calls rapidly, sometimes dozens of times before a user finishes reading the early parts of its output. Prompt caching is another key optimization, often reaching over 90 percent cache hit rates in partner environments. Benchmarks show grok-code-fast-1 producing responses at 190 tokens per second, while maintaining a significantly lower price point than other coding-focused models.
Programming versatility
The model is versatile across the full development stack. It performs well in languages such as TypeScript, Python, Java, Rust, C++, and Go. In practice it can handle a wide range of tasks, from building new applications to answering questions about large codebases, fixing bugs, and making precise code edits. Developers who tested the model report that its speed changes the way they use coding assistants, allowing them to work more fluidly.
Pricing and accessibility
xAI has set pricing for grok-code-fast-1 at $0.20 per million input tokens, $1.50 per million output tokens, and $0.02 per million cached input tokens. This makes the model cost-effective while keeping performance high, positioning it as a daily driver for software engineers.
Evaluation and methodology
Performance was measured through a blend of benchmarks and real-world tests. On the SWE-Bench-Verified dataset, grok-code-fast-1 reached 70.8 percent accuracy. The xAI team emphasized that human evaluations by experienced developers were equally important for shaping the model’s capabilities, particularly in measuring real usability during end-to-end coding workflows.
Looking ahead
The model was quietly released last week under the codename sonic, allowing the team to monitor community feedback and push updates during a stealth phase. xAI plans to improve the model continuously, releasing updates on a rapid cycle measured in days. A new version currently in training will add multimodal support, parallel tool execution, and longer context windows.
grok-code-fast-1 is positioned as a practical choice for developers who want a reliable and economical coding assistant that can keep pace with the demands of real-world software engineering.
Add a comment
Tags: ML News
Iterate on AI agents and models faster. Try Weights & Biases today.