Grok 4.7 发布

SpaceX AI推出Grok 4.7模型，将其定位为最强大的编码和知识工作AI模型1。该模型在长任务处理、代码生成和自我检查能力方面相比Grok 4.6有显著提升1，并在CursorBench 4.0上处于价格-性能前沿1。Grok 4.7采用更大的基础模型架构，经过更长时间和更难的强化学习训练1。

在安全防护方面，Grok 4.7在生物安全基准测试上达到62.4%1，在HackerBench v0.3上仅允许3.3%的危险双用途提示通过1，展现业界领先的安全防护能力1。定价方面，输入tokens为每百万$2，输出tokens为每百万$61。此外还提供快速版本，输出速度提升两倍，价格相应翻倍1。

Grok 4.7已在Cursor、Grok Build、Grok API及第三方平台上线1。

SpaceX AI has unveiled Grok 4.7, a large language model designed as the most powerful AI system for coding and knowledge work tasks 1. The model delivers substantial improvements over its predecessor, Grok 4.6, particularly in long-task handling, code generation, and self-correction capabilities, while maintaining the same pricing and speed as before 1.

Built on a larger foundational model and trained with extended reinforcement learning processes and more challenging datasets, Grok 4.7 achieves leading price-to-performance efficiency on CursorBench 4.0 1. The system demonstrates industry-leading safety protections, scoring 62.4% on LatchBio's biosafety benchmark and allowing only 3.3% of dangerous dual-use prompts to pass through on HackerBench v0.3 1. Pricing remains at $2 per million input tokens and $6 per million output tokens, with a faster variant offering twice the output speed at double the cost 1. The model is now available across multiple platforms, including Cursor, Grok Build, the Grok API, and third-party services 1.