趋近智
技术规格
注意力结构
Multi-Head Attention
隐藏维度大小
-
层数
-
注意力头
-
键值头
-
激活函数
-
归一化
-
位置嵌入
Absolute Position Embedding
Grok 4.1
Grok 4.1 brings significant improvements to real-world usability with exceptional creative, emotional, and collaborative capabilities. Optimized for style, personality, helpfulness, and alignment using frontier agentic reasoning models as reward models. Achieves #1 on LMArena Text Leaderboard with 1483 Elo (thinking mode) and #2 with 1465 Elo (non-thinking), surpassing all other models. Features 2M context window, reduced hallucination rate (12.09% → 4.22% on production queries), and state-of-the-art emotional intelligence (1586 Elo on EQ-Bench). Available in both reasoning and fast non-reasoning modes through API.
关于 Grok 4
xAI's frontier intelligence models trained with reinforcement learning at unprecedented scale using the 200,000 GPU Colossus cluster. Grok 4 series demonstrates state-of-the-art performance in reasoning, coding, and multimodal understanding with native tool use capabilities. Features real-time search integration across X and the web, advanced reasoning through scaled RL training, and industry-leading performance on academic benchmarks. Designed for both immediate responses and extended thinking modes with vision capabilities.
其他 Grok 4 模型
评估基准
没有可用的 Grok 4.1 评估基准。