xAI's Grok 4.5 targets coding and agents, undercuts rivals on token price

xAI released Grok 4.5 on July 8, 2026, positioning it as its most capable model yet for software engineering, agentic workflows, and general knowledge work. The company says it was co-developed with Cursor and trained across tens of thousands of NVIDIA GB300 GPUs, with heavy emphasis on data curation and a reinforcement-learning stack tuned for long-running, asynchronous agentic rollouts spanning hundreds of thousands of multi-step technical tasks. It’s now the default in Grok Build and ships with Word, PowerPoint, and Excel plugins that can do web research, build multi-sheet spreadsheet models, and assemble slide diagrams.

The pitch centers on cost and speed rather than raw benchmark bragging. Grok 4.5 runs at roughly 80 tokens per second and is priced at $2 per million input tokens and $6 per million output tokens, with xAI claiming about twice the token efficiency of comparable frontier models — meaning fewer steps to finish a task. Those efficiency and ‘smartest model’ figures are vendor claims, not independently verified, and the demos (a Three.js solar-system simulation, auto-generated Office documents) are cherry-picked single-prompt outputs.

Availability is immediate in Grok Build, in Cursor across all plans, and via the xAI console API, with free usage offered for a limited time. The notable gap: the EU is excluded at launch across all xAI products and the API, with access expected in mid-July — a rollout delay that continues the pattern of frontier models shipping late in Europe over regulatory friction.