Grok Code Fast 1: xAI’s New Coding Model 3× Faster, 6× Cheaper

On Thursday, Elon Musk’s xAI officially launched its latest code model, Grok Code Fast 1, arriving just before Musk’s promised August deadline.

The model is considered the code‑focused version of Grok 4, designed to provide a fast and cost‑effective solution for “agentic programming,” where AI automatically invokes tools such as grep, terminals, and file editors within an IDE to complete coding tasks.

xAI notes that while today’s large language models are powerful, they are not specifically built for agentic coding workflows, prompting engineers to create a more flexible, faster‑responding solution optimized for everyday tasks.

Grok‑code‑fast‑1 is a ground‑up trained language model using a new architecture; xAI built a rich programming‑focused pre‑training corpus and selected high‑quality datasets that reflect real‑world pull requests and coding tasks.

During training, xAI collaborated closely with partner platforms to refine model behavior; grok‑code‑fast‑1 has mastered common tools like grep, terminals, and file editing, enabling easy adoption in typical IDEs.

At launch, xAI announced a free one‑week trial of grok‑code‑fast‑1 on many platforms, including GitHub Copilot, Cursor, Cline, Roo Code, Kilo Code, opencode, and Windsurf. Earlier that week, the model had silently rolled out on some platforms under the codename Sonic.

The blog post and model card outline some features, though details on architecture, data, and fine‑tuning are sparse; xAI’s inference and super‑computing teams introduced innovations that dramatically boost service speed, delivering a responsive experience where the model invokes dozens of tools before a user finishes reading the first line of AI reasoning.

xAI also invested heavily in fast cache optimization, achieving cache hit rates typically above 90% across partner platforms.

Grok‑code‑fast‑1 is highly flexible across the software development stack, excelling in TypeScript, Python, Java, Rust, C++, and Go, and can perform common coding tasks with minimal supervision—from building projects from scratch and offering deep codebase insights to executing precise bug fixes.

For example, using grok‑code‑fast‑1, Danny Limanseta created a simple game in a single day:

The model’s pricing is relatively low:

Input tokens: $0.20 per million

Output tokens: $1.50 per million

Cached input tokens: $0.02 per million

Designed for developers’ everyday tasks, it balances performance and cost, making it a versatile choice for fast, efficient handling of common coding work.

In the SWE‑Bench‑Verified subset test, grok‑code‑fast‑1 achieved a 70.8% score, approaching the leading Claude 4 series. xAI emphasizes that development was guided by real‑world human evaluation, focusing on usability and user satisfaction, and many programmers now rate the Grok model as a fast, reliable tool for daily coding tasks.

xAI says its team will continue updating grok‑code‑fast‑1, with a new variant supporting multimodal input, parallel tool calls, and extended context length currently in training.