xAI Launches Grok 4.5: Cursor-Trained Coding Model with 2x Token Efficiency

xAI

Models / LLM official + media 3 src. ~1 min

xAI released Grok 4.5 on July 8–9, co-trained with Cursor and targeting coding, agentic tasks, and knowledge work. xAI claims roughly 2x token efficiency over comparable leading models (tasks solved in under half the steps), served at 80 tokens per second. Pricing: $2/M input, $6/M output. Available in Grok Build, Cursor (all plans), and the xAI API console; EU availability expected mid-July.

Why it matters

Grok 4.5's token efficiency claim challenges the assumption that frontier capability requires proportionally high compute per task. The Cursor co-training partnership signals a new pattern where AI labs and developer tooling companies jointly optimize models for real engineering workflows.

Importance: 4/5

xAI frontier model release co-trained with Cursor; Opus-class positioning with 2x token efficiency claims.

Sources