xAI has released Grok 4.5, a model trained on tens of thousands of Nvidia GB300 GPUs targeting coding, agentic tasks, and knowledge work. The release includes updates from Artificial Analysis ranking the model fourth on its Intelligence Index behind Fable 5, GPT-5.5, and Opus 4.8.

  • Grok 4.5 costs $2 per million input tokens and $6 per million output tokens, significantly undercutting competitors like Fable 5 ($10/$50) and Opus 4.8 ($5/$25).
  • On Terminal Bench 2.1, it scores 83.3%, trailing Fable 5 by one point, but falls behind GPT-5.5 (67%) and Fable 5 (70%) on DeepSWE 1.1.
  • Artificial Analysis notes a hallucination rate jump to 54% despite accuracy improvements, though it performs well on agentic tasks in Grok Build.
  • The model is available via Grok Build, Cursor, and the xAI console, with EU availability targeted for mid-July.

xAI positions Grok 4.5 as a cost-efficient alternative to frontier models by leveraging heavy data filtering and asynchronous learning infrastructure to reduce token usage and price.