Grok-4

model
xAI

MoE architecture with sparse + dilated attention. 256K token context. Trained with 100x more compute than Grok-2 and 10x more RL compute than Grok-3. Native tool-calling, real-time search, and extended reasoning.

Grok-4 Fast variant offers 2M context at $0.20/$0.50 per million tokens (AA index 35). AA Intelligence Index: 42. Proprietary.

AnnouncementArtificial AnalysisOpenRouter

Model Details

Architecture MOE
Context window 256,000

Variants

Name Parameters Notes
Grok-4 — 256K context
Grok-4 Fast — 2M context, AA index 35
frontierreasoningagentsmoe

Related