Grok 4.5 comes out of hiding with a $2 price tag and third-party numbers
Two weeks after lvl30 called Grok 4.5's private-beta claims unverifiable, there are checkable numbers: $2 per million input tokens, behind Fable 5 and GPT-5.5 on coding, far fewer tokens burned.
Two weeks ago I covered Grok 4.5 as a teaser with a parameter count attached: a private beta at SpaceX and Tesla with no public access and no independent numbers. Credit where due, the situation improved. Per The Decoder, Grok 4.5 now has public pricing at $2 per million input tokens, and third-party benchmark analysis that puts it behind Claude Fable 5 and GPT-5.5 on coding while burning about 4.2 times fewer tokens than Opus 4.8 on comparable tasks. xAI says it trained the model on tens of thousands of NVIDIA GB300s; EU availability is expected mid-July.
The token-efficiency angle is the interesting one. Per-token prices stopped meaning much once reasoning models started varying their token spend by an order of magnitude per task, so a model that is slower on the leaderboard but frugal at the meter can genuinely come out cheaper where it counts. That is the same lesson the GPT-5.6 pricing tells from the other direction.
The June take here stands: a model counts when there is something to run, an API to hit, and numbers produced by someone other than the vendor. Grok 4.5 now clears that bar. What it does on real work is a different question, and that one still gets answered on the bench, not in a launch chart.