Grok代码模型来了：限时免费用，速度超级快

Core Viewpoint - xAI has launched Grok Code Fast 1, a new code model that is three times faster than GPT-5 and six times cheaper, aimed at providing efficient solutions for agentic programming tasks [2][4]. Group 1: Model Features - Grok Code Fast 1 is designed for AI to automatically execute programming tasks within IDEs, utilizing tools like grep and terminal [4]. - The model is trained from scratch with a new architecture and a pre-training corpus rich in programming content, ensuring it can handle real-world coding tasks effectively [4][6]. - It has a high cache hit rate of over 90% when running on partner platforms, enhancing its performance [7]. Group 2: Performance and Cost - Grok Code Fast 1 scored 70.8% on the SWE-Bench-Verified subset tests, positioning it close to the top-performing models like Claude 4 [10][12]. - The pricing structure is competitive, with costs of $0.20 per million input tokens, $1.50 per million output tokens, and $0.02 per million cached input tokens [10]. Group 3: Future Developments - xAI plans to continuously update Grok Code Fast 1, with a new variant in training that will support multimodal inputs and parallel tool calls [13].