Cursor和SpaceXAI联手推出Grok 4.5，能处理编码、数据、金融等复杂任务，比Composer 2.5更通用，首周双倍用量。
Grok 4.5由Cursor与SpaceXAI联合训练，采用混合专家架构，基于数万亿token训练，包含真实Cursor用户交互及STEM、科研论文等数据。强化学习聚焦长任务、工具使用、错误恢复和验证，覆盖编码、数据、金融等领域。定价为输入每百万token 2美元、输出6美元，同时保留更快的变体。模型已在桌面应用、Web、iOS、CLI和SDK上线，首周付费计划双倍用量。欧盟暂不可用，预计本月晚些时候支持。
Cursor and SpaceXAI has released Grok 4.5 today af…
Cursor and SpaceXAI has released Grok 4.5 today after joint training.
It is a mixture-of-experts model trained on trillions of tokens from real Cursor user interactions plus broader STEM, research papers and knowledge work data. Reinforcement learning focused on long, messy tasks that need tool use, error recovery and verification across coding, data work, finance and similar areas.
The base pricing sits at $2 per million input tokens and $6 per million output, with a faster variant at higher rates. It sits in a heavier weight class than Composer 2.5, which stays available as the dedicated coding model with new versions planned. This gives Cursor users a wider general tool rather than another pure coding specialist, though how well it holds up on complex non-coding workflows outside their own environment is the bit worth watching in practice.
It is available right now in the desktop app, web, iOS, CLI and SDK, with double usage included for the first week on paid plans.
Grok 4.5 isn't available in the EU yet. EU availability is expected later this month. https://t.co/q59hIpM9w9