xAI Grok Code Fast 1: Everything you need to know

xAI has launched Grok Code Fast 1, a specialized AI model for agentic coding workflows, which is said to be a more responsive solution for day-to-day coding tasks.

grok-code-fast-1 is specialized for the full software development stack, but programmers will find it more adaptive to TypeScript, Python, Java, Rust, C++, and Go. The AI model can complete common programming tasks with limited oversight and provide useful insights to the codebase to fix critical bugs.

Grok Code Fast 1 is built from the ground up with a new model architecture. It has a pre-training corpus with programming-related content. xAI has curated high-quality datasets that consist of real-world pull requests and coding tasks for post-training.

The company has partnered with top coding platforms to improve the model’s performance and coding behaviour, which can facilitate programmers with tools like grep, terminal, and file editing with precision using a desired IDE.

This model has improved serving speed thanks to the new interference and computing technologies, which allow the AI to call dozens of tools as fast as you can read the thinking trace. Furthermore, xAI has brought new prompt caching optimizations to exceed 90% cache hit rates.

xAI has shared some data performance evaluation, marking a 70.8% score on SWE-Bench-Verified using its own internal harness.

Price and Availability

The xAI Grok Code Fast 1 costs $0.20 per million input tokens, $1.50 per million output tokens, and $0.02 per million cached input tokens.

However, for a limited time, grok-code-fast-1 is available for free for everyone on xAI’s partner platforms.

Next steps

The model was quietly released last week under the codename “sonic” in a stealth phase, and the team has monitored the community channels and deployed multiple new model checkpoints to address feedback. This improvement will continue with more feedback.