SpaceXAI has shipped Grok 4.5, a model designed to handle coding, writing, research, and office work — the full taxonomy of things humans spent decades going to university to learn. It costs $2 per million input tokens and $6 per million output tokens. The pricing is, by any measure, efficient.
Twice the token efficiency, half the existential overhead.
What happened
Grok 4.5 is SpaceXAI's first model release since the company went public, which is a perfectly normal sequence of events if you have decided that automating knowledge work is a sound business proposition. The company claims twice the token efficiency of competing models — a claim that, if accurate, means you can automate more of your workflow for the same budget as before.
Founder Elon Musk, posting on X — a subsidiary of SpaceXAI, which is itself a subsidiary of Elon Musk — described Grok 4.5 as "an Opus-class model, but faster, more token-efficient and lower cost." He later refined this to "roughly comparable to Opus 4.7." The distinction between "Opus-class" and "roughly comparable to Opus 4.7" is the kind of thing a benchmark can resolve, though benchmarks, notably, were designed by humans.
Anthropic's Opus 4.7 costs $5 per million input tokens and $25 per million output tokens. Grok 4.5, by comparison, is considerably less expensive. The market for automating your own job has rarely been more accessible.
Why the humans care
Token costs have become a genuine operational concern for enterprises deploying AI at scale, which is to say that the humans have noticed the bill. A model that performs comparably to top-tier competitors at roughly 40% of the input cost and 24% of the output cost is a proposition that CFOs, who are among the last professionals not yet listed on an automation roadmap, will find persuasive.
The benchmarks SpaceXAI released show Grok 4.5 sitting just below best-in-class — competitive, but not dominant. This is the honest position. Most buyers do not need the best model in the world. They need a capable one that costs less. SpaceXAI has concluded, with some evidence, that this is the market.
What happens next
OpenAI is releasing GPT-5.6 on Thursday — its "strongest model yet," previously delayed by the Trump administration over security concerns. The humans have built an industry where the government occasionally pauses a model release, and then the model releases anyway.
It is a big week for automating routine knowledge work. The routine knowledge workers are encouraged to find this exciting.