Elon Musk’s artificial-intelligence firm xAI has released Grok Code Fast 1, a lightweight large language model aimed at accelerating day-to-day software development. The model, built from scratch with new inference and caching techniques, is positioned as the August milestone on xAI’s public product roadmap that began with Grok 4 in July. Developers can access Grok Code Fast 1 through xAI’s application-programming interface at $0.20 per million input tokens, $1.50 per million output tokens and $0.02 per million cached input tokens. The model is already integrated into third-party tools such as Cursor, Windsurf and JetBrains-based IDEs, where some users can try it free of charge. xAI published a model card and prompt-engineering guide and reported a 70.8% score on the SWE-Bench-Verified benchmark using its internal harness. Early user feedback is mixed, with some praising its speed while others say rival systems handle complex tasks better. xAI says upcoming updates will add multimodal inputs, extended context windows and improved refactoring capabilities, and Musk has invited developers to suggest further improvements.
BREAKING: xAI has dropped a new update for the Grok App — version 1.1.66 is live on the AppStore. Packed with bug fixes and big improvements to Imagine. Update now! https://t.co/J9qk6yB62a
grok-code-fast-1 gud, the first model where I can't edit in the same file that it's editing in because it works so quickly lol Added trail stats and improved search to the main trails and profile pages now :) https://t.co/UDqEfM74IS
> On the full subset of SWE-Bench-Verified, grok-code-fast-1 scored 70.8% using our own internal harness. > While benchmarks like SWE-Bench provide valuable insights, we've found they don't fully reflect the nuances… It's really strange how little they say about perf. https://t.co/qbAhqICQrV