SpaceXAI 发布 Grok 4.6 模型，主打长程 agent 任务
Grok 4.6 以低价和高跑分重回第一梯队，同时 Grok Bot 主打长程自主执行，背后是 SpaceXAI 与 Cursor 的深度整合。
Grok 4.6 跑分反超 GPT-5.6 Sol 和 Fable 5 Max，但价格更低。它擅长知识工作类的 agentic 任务，在纯终端操作上仍落后。新模型已接入 Cursor 等平台，并同步发布了可 24 小时运行的 Grok Bot。
正文摘录
- title: Musk's Grok 4.6 Back in the Top Tier! Lower Price Beats Fable 5 — That Cursor Acquisition Wasn't for Nothing - sourcecompany: QbitAI - bodymarkdown: Its benchmark scores beat GPT-5.6 Sol and Fable 5 Max, yet the price is only $2 per million input tokens and $6 per million output tokens — considerably cheaper than both rivals. On GDPVal-AA v2, which measures real-world work capability, Grok 4.6 achieved the highest score of the field at 1753, leaving the others behind. It also ranked first on AA-Briefcase and Harvey LAB. The new model is now integrated into Grok Build, Cursor, Grok Bot, and the API, with double usage available for the first week in Cursor. The focus of this upgrade is long-horizon agentic tasks — the model needs to work continuously for extended periods without hum…