Just a couple of weeks after SpaceX released its Grok 4.7 model and claimed that it was competitive with frontier models in coding and knowledge work, the company is now letting those models do Grok’s work. CEO Elon Musk announced via Twitter that the Grok Bot agent service will now select from a range of different models before choosing “whatever is most likely to give you the best outcome” for a user’s request.
Per Musk, available models will include Anthropic’s Claude Opus 5.5, AI image generator MidJourney, AI audio maker Suno, and presumably others that weren’t named directly. SpaceX’s own model, Grok 4.7, is also presumably one of the options, though Musk did not specify. “Simple questions will route to small, fast models. Questions with complex answers will route to large models,” he explained.
There is no indication that SpaceX won’t continue to develop its own model—Musk was literally predicting that it would bypass all competitors by version 5 just a few weeks ago. He’s also reportedly changing the name of SpaceXAI to SpaceXSI to reflect the company’s pursuit of “super intelligence” (and to brown-nose Donald Trump, who is still trying to give AI a rebrand to SuperIntelligence to improve public perception).
But allowing the Grok AI agent to access other models on the backend does suggest that SpaceX is continuing to look at another avenue other than frontier model domination. Instead, the company could be looking to serve as the infrastructure that supports other AI companies that do the significantly more expensive work of development.
Earlier this year, SpaceX entered an agreement with rival Anthropic, a company Musk once called “evil,” to let the company use SpaceX’s massive Colossus data center to train and run its Claude model. It seems to have rented space to other companies, too. On Twitter, Musk confirmed that the new open-source model called Beam, developed by Reflection AI, was trained on Colossus servers.
It’d be a pivot that would make sense, given the sheer amount of money that SpaceX has already spent building out its AI infrastructure. There is a much cleaner business model in renting out server space than there is in trying to figure out how to monetize an AI model, no matter how powerful it may be. Plus, Grok does not exactly have a great reputation on its own. Maybe letting it speak Claude to people (you know, instead of declaring itself MechaHitler) would improve its reputation, but it’d almost certainly be cheaper to let Anthropic do all the computing instead of running everything through the Grok model.