Musk's programming model is available for free for a limited time. It has a fast response speed and challenges GPT-5.

According to Zhidx on August 29th, last night, Elon Musk's AI company, xAI, launched the reasoning model Grok Code Fast 1 that excels in programming. The API call name is grok - code - fast - 1.

Elon Musk announced the launch of the programming model

Grok Code Fast 1 supports a context window length of 256K and features fast response and cost - effective pricing. Before users finish reading the first paragraph of the thinking trajectory, the model has already invoked dozens of tools. xAI has also invested in fast cache optimization, with cache hit rates usually exceeding 90%.

The pricing of the model is as follows: $0.2 (approximately 1.43 Chinese Yuan) per million input tokens, $1.5 (approximately 10.70 Chinese Yuan) per million output tokens, and $0.02 (approximately 0.14 Chinese Yuan) per million cached input tokens.

As shown in the official diagram released by xAI, the performance of Grok Code Fast 1 is comparable to that of models such as GPT - 5 and Gemini 2.5 Pro, but its price is several times lower. Its price is slightly higher than that of Qwen3 - Coder, but it has a higher number of tokens per second.

Comparison of Grok Code Fast 1 with models such as GPT - 5, Gemini 2.5 Pro, and Qwen3 - Coder

It is reported that Grok Code Fast 1 is particularly proficient in TypeScript, Python, Java, Rust, C++, and Go. It can complete common programming tasks with minimal supervision, ranging from building projects from scratch, providing in - depth solutions to codebase problems, to performing precise error repairs.

For example, within less than a day, users can use Grok Code Fast 1 on Cursor to build the prototype of a battle simulator. Users can iterate quickly and control it precisely to achieve the desired effect.

Another example is that within less than a minute, users can complete the process from concept to model design. Its mobile - responsive design is also excellent. The UI it generates looks good on both desktop and mobile devices without much adjustment.

xAI built Grok Code Fast 1 from scratch using a brand - new model architecture. It constructed a pre - trained corpus rich in programming - related content and selected a high - quality dataset that reflects real - world pull requests and programming tasks during the post - training phase. Eventually, Grok Code Fast 1 mastered the use of common tools such as grep, terminals, and file editing.

xAI uses a holistic approach to evaluate model performance, combining public benchmarks with practical tests. In the full subset test of SWE - Bench - Verified, using its own internal testing tools, Grok Code Fast 1 scored as high as 70.8%.

The overall results of Grok Code Fast 1 in ToyBench currently rank 5th, behind GPT - 5, Claude Opus 4, Gemini 2.5 Pro, and DeepSeek Reasoner.

Currently, Grok Code Fast 1 is available for free use for a limited time of 7 days and can be accessed on GitHub Copilot, Cursor, Cline, Roo Code, Kilo Code, opencode, and Windsurf.

This article is from the WeChat official account “Zhidx” (ID: zhidxcom). The author is Li Shuiqing, and the editor is Xin Yuan. It is published by 36Kr with authorization.