Today, xAI announced a new AI model focused on agentic coding workflows. This new grok-code-fast-1 model is built from the ground up on a brand-new model architecture with more programming-related content during pre-training. During post-training, xAI used high-quality datasets that reflect real-world pull requests and coding tasks to make the model more suitable for day-to-day software engineering tasks.
We are thrilled to announce the release of our 1st model for coding and agentic tasks, grok-code-fast! You can now integrate it into all IDE and CLI tools for your daily coding needs. It’s fast, powerful, and affordable! We’re committed to continuous improvement based on your… https://t.co/YcJx8ROfHJ
— Bill Yuchen Lin (@billyuchenlin) August 28, 2025
This new grok-code-fast-1 model is optimized for TypeScript, Python, Java, Rust, C++, and Go languages, and it is now available for free for a limited time on popular vibe-coding platforms including GitHub Copilot, Cursor, Cline, Roo Code, Kilo Code, OpenCode, and Windsurf.
The main highlights of this new coding model are its pricing, speed, and performance. Here"s the pricing of this new model:
- $0.20 per million input tokens
- $1.50 per million output tokens
- $0.02 per million cached input tokens
xAI mentioned that this model scored 70.8% on the full subset of SWE-Bench-Verified using their own internal harness. It remains to be seen how this model fares in independent tests in the coming weeks.
The xAI team wrote the following regarding the model"s performance:
"When developing grok-code-fast-1, we focused on usability and user satisfaction, guided by real-world human evaluations. The result is a model rated by programmers as fast and reliable for everyday coding tasks."
Another main highlight of grok-code-fast-1 is its speed. xAI"s inference and supercomputing teams used several new techniques to significantly accelerate token generation speed up to 160 tokens per second. They have also included prompt caching optimizations, which now achieve hit rates above 90% when used with GitHub Copilot, Cursor, and other launch partners.