Company xAI Elon Musk introduced Grok Code Fast 1 — a new AI model focused on agent-based coding (working with multi-step coding and tools).
Grok Code Fast 1 was created from scratch using a completely new architecture. At the pre-training stage, we collected a data corpus as rich as possible in programming materials. For post-training, we selected datasets that reflect real-world scenarios: pool requisitions, work tasks with code, and typical changes in projects. Throughout the training, the team collaborated with the launch partners to hone the model’s behavior in agent platforms. Thanks to this, Grok Code Fast 1 is well versed in working with common tools — grep, terminal, file editing — and feels “at home” in most IDEs.
Together with the launch partners, the model is currently available for free for a limited time in the following products: GitHub Copilot, Cursor, Cline, Roo Code, Kilo Code, opencode, and Windsurf.
High speed performance
When developing the model, special attention was paid to its speed. Thanks to this, it can manage to make dozens of tool calls before the user finishes reading the first paragraph of the “chain of thought”. Additionally, optimizations have been implemented that provide more than 90% of cache hits when working with launch partners.
The Grok Code Fast 1 model works across the entire development stack, but is especially good at TypeScript, Python, Java, Rust, C++, and Go.
The model is capable of performing typical tasks with little or no supervision:
- creating projects from scratch;
- answers to questions about code and architecture;
- point bug fixes.
The new model was tested both in general tests and provided for evaluation by real developers who tested it in their daily tasks. According to the company, Grok Code Fast 1 has been highly praised by programmers as a fast and reliable model for everyday tasks.
Price
Grok Code Fast 1 is designed with a focus on widespread use. For a limited period, it is available for free from exclusive launch partners. It is worth mentioning that it was previously released “quietly” under the codename sonic.
In the future, the following tariffs have been announced for it:
- 0.20 per 1 million incoming tokens,
- $1.50 for 1 million initial tokens,
- 0.02 per 1 million cached incoming tokens.
Source: x.ai
Spelling error report
The following text will be sent to our editors: