SpaceX AI 今日正式推出了其最新且性能最强的语言模型 Grok 4.5 [1]。该模型专为编码、智能体（agentic）任务及知识工作场景进行了优化设计 [1]。在代码生成和复杂工程任务方面，Grok 4.5 展现了卓越的表现能力 [1]。
为了提升模型的推理性能，其训练过程动用了数万块 NVIDIA GB300 GPU，并采用了强化学习技术进行增强 [1]。目前，该模型已在 Grok Build、Cursor 编辑器以及 API 控制台正式上线并提供限时免费使用 [1]。对于开发者而言，Grok 4.5 默认集成于 Cursor 编辑器的所有订阅计划中 [1]。此外，用户还可以通过 Microsoft Word、PowerPoint 和 Excel 的插件调用该模型服务 [1]。
在定价方面，输入 token 的费用为每百万个单位 2 美元，输出 token 则为每百万个单位 6 美元 [1]。针对欧盟地区的用户，Grok 4.5 预计将于今年 7 月中旬开放访问 [1]。
SpaceXAI has officially launched its latest large language model, Grok 4.5, which the company describes as its strongest to date [1]. The new model is specifically optimized for coding, agentic tasks, and knowledge work [1]. It demonstrates exceptional performance in code generation and complex engineering tasks [1].
To achieve these capabilities, the training process utilized tens of thousands of NVIDIA GB300 GPUs and employed reinforcement learning techniques to enhance reasoning abilities [1]. Grok 4.5 is now available on the Grok Build platform, via API consoles, and within Cursor editor with a limited-time free offer currently active for users [1].
The model integrates deeply into development workflows through default availability in all plans of the Cursor editor [1]. Additionally, it supports plugins for Microsoft Word, PowerPoint, and Excel [1]. Pricing is set at $2 per million input tokens and $6 per million output tokens [1]. Availability for users in the European Union region is expected to open by mid-July [1].