SpaceXAI Releases Grok 4.7, "Twice as Fast at Half the Price" of a Model It Would Rather Not Name
The model is also served at exactly the same price and speed as Grok 4.6, a fact the same announcement describes as "highly competitive."
AUSTIN — SpaceXAI on Monday released Grok 4.7, which the company describes in the first sentence of its announcement as "twice as fast, at half the price of comparable models," and in the fourth sentence as "served at the same price and speed as Grok 4.6."
"Both statements are true," said a company spokesperson. "It is twice as fast as a comparable model, and it is the same speed as our previous model. We would ask you to consider what that says about the comparable model."
The announcement includes a table of seven benchmarks in which Grok 4.7 is compared against three other models, one of which is Grok 4.6. Grok 4.7 beats Grok 4.6 on all seven.
"That is the frontier," the spokesperson said. "We define the frontier as the area in front of where we were."
Among the model's improvements, the company said, is that it was "trained to natively understand the Grok Bot harness," meaning it is now the best model in the world at being run by the thing that runs it. It also "works longer on difficult tasks," a capability the release does not compare against anyone.
"Every lab does this," said Anneliese Vogt, an analyst who tracks model launches. "You pick the benchmarks where you win and you pick the neighbors who are cheaper than you. The difference is most labs spread it over a few paragraphs. This one put 'half the price' and 'same price' within 60 words of each other and left a chart in between as a kind of buffer."
The company said the model scored 19.6 percent on a legal-work benchmark, a result it described as leading its class, and which a paralegal contacted for comment described as "a paralegal."
At press time, the company had described Grok 4.6 as "comparable."