Grok Code Fast 1 release notes / what's new

xAI has announced the release of grok-code-fast-1, a model built from the ground up with a new architecture specifically designed for agentic coding workflows. The model aims to reduce the latency often associated with reasoning loops and tool calls, providing a more responsive experience for developers.

High-Speed Inference and Agentic Optimization

grok-code-fast-1 is engineered for extreme responsiveness, enabling the model to execute multiple tool calls before a user finishes reading the initial thinking trace. This performance is driven by innovations from xAI's inference and supercomputing teams, complemented by prompt caching optimizations that regularly achieve hit rates exceeding 90% when used with launch partners.

To support agentic workflows, the model was trained to master common developer tools, including:

File editing
Terminal access
Grep

Technical Foundation and Capabilities

The model was developed using a brand-new architecture and a pre-training corpus rich in programming content. Post-training involved curated datasets reflecting real-world coding tasks and pull requests.

grok-code-fast-1 is versatile across the software development stack, with specific proficiency in the following languages:

TypeScript
Python
Java
Rust
C++
Go

Its capabilities range from creating "zero-to-one" projects and answering codebase questions to performing surgical bug fixes.

Performance Benchmarks and Evaluation

On the full subset of SWE-bench Verified, grok-code-fast-1 achieved a score of 70.8% using xAI's internal harness.

Beyond public benchmarks, xAI utilizes a holistic evaluation approach combining:

Routine human assessments from experienced developers rating end-to-end performance on everyday tasks.
Automated evaluations to track behavioral trade-offs.
Real-world testing to ensure usability and user satisfaction.

Pricing and Accessibility

grok-code-fast-1 is positioned as an economical choice for daily coding tasks, with the following API pricing:

Input tokens: $0.20 per million
Output tokens: $1.50 per million
Cached input tokens: $0.02 per million

The model is generally available via the xAI API. Additionally, it is offered for free for a limited time through select launch partners, including GitHub Copilot, Cursor, Cline, Roo Code, Kilo Code, opencode, and Windsurf.

"In early testing, Grok Code Fast has shown both its speed and quality in agentic coding tasks. Empowering developers with powerful tools is a core part of our mission at GitHub Copilot, and this is a compelling new option for our developers." — Mario Rodriguez, Chief Product Officer, GitHub

Future Roadmap

Following a stealth release under the codename sonic, xAI plans to iterate rapidly on the model based on community feedback. A new variant is currently in training that will introduce:

Multimodal input support
Parallel tool calling
Extended context length