Grok 4.7 在 AI 编程代理测试中击败 GPT-5.6 Sol

🚀 Grok 4.7 beats GPT-5.6 Sol in tests of AI programming agents According to Artificial Analysis, the new SpaceXAI model scored 56 points in the Coding Agent Index - 9 more than Grok 4.6. In conjunction with Grok Build, it took fourth place among models in native tools, behind only Claude Fable 5.1, GPT-6 Astra and Claude Opus 5. Grok 4.7 was especially noticeably improved for long agent tasks: on the closed AA-Briefcase, the result immediately increased by 111 Elo, bringing the model closer to the leaders of Claude. But in xhigh mode, Grok 4.7 spends on average about 81 thousand output tokens per task - versus 36 thousand for Grok 4.6 and 27 thousand for GPT-6 Astra. [Source]()