SpaceXAI unveils Grok 4.5, a flagship model at 2/6 dollars per million tokens

On July 8 SpaceXAI (Elon Musk's AI company, heir to xAI) unveiled Grok 4.5, its most capable model, opened to the public the following day. The positioning does not chase absolute benchmark supremacy: the company itself concedes results just short of best-in-class. It chases your budget instead.

Musk's chosen comparison is blunt: an Opus-class model (the reference is Anthropic's Claude Opus 4.7, considered the standard for high-end agentic work), but faster, more efficient, and much cheaper. The price list is the message: 2 dollars per million input tokens and 6 for output, against 5/25 for Opus 4.7 and 5/30 for OpenAI's new Sol.

Grok 4.5 is described as a workhorse: writing and fixing code, building small applications, office and clerical tasks, research and writing. The idea is to cover the everyday reality of white-collar and developer work, where cost and speed matter more than winning any single test.

For the market it is another salvo in the price war running through the summer of 2026: if a "nearly top-tier" model costs a third to a fifth of its direct rivals, pressure grows on Anthropic and OpenAI to justify their price lists, and for AI application builders the cost per completed task becomes the real battleground.