Musk says SpaceX's Grok bot will route tasks to rivals
The company with its own frontier lab is now buying competitors' models task by task.
Elon Musk says SpaceX will route tasks on its Grok bot to whichever model does the job best, naming Claude Opus 5.5, MidJourney, Suno and other leading APIs. He framed it as an "Important note regarding Grok" and ended with the reasoning, "Whatever is most likely to give you the best outcome."
@SpaceX will use the best back end model for any given task, including Claude Opus 5.5, MidJourney, Suno and other leading APIs.
That is the whole announcement. No routing logic, no cost figures, no quality comparisons, no timeline. Nobody outside the company knows which tasks go where, or how often the bot will reach for an outside API rather than the in-house model.
Matt Shumer, who has been vocal about model choice, read it as a straight win for people using the bot.
The business read
The interesting part is who is doing it. This is an organisation in the orbit of its own frontier lab, and it is still shopping per task across image, audio and reasoning providers rather than defaulting to the house model everywhere. If that is the calculation when you own a model, the argument for your own product pledging loyalty to one vendor gets thinner.
For a small software business the practical version is the same shape. You pick a model for the job, you swap when something better lands, and you do not sell your users on the brand of the engine. Most of the builders we follow have been drifting that way already, mixing providers by task and by price. Musk naming three competitors in one post makes the pattern harder to call a nerd habit.
What it does not tell you is whether this is cheaper or better in practice. Routing across providers adds its own work, from fallbacks to latency to keeping output consistent when the thing behind the curtain changes. None of that is addressed in the post, and no numbers were published alongside it.
Worth watching for two things. First, whether SpaceX publishes anything concrete about how the routing picks a model, because that is the part other builders could actually copy. Second, whether users notice the difference at all, which is the only test that matters for a bot people talk to every day.
For now it is one sentence of policy from Musk and one line of approval from Shumer. Treat it as a signal about how even the best-resourced teams are thinking about model choice, not as evidence that anything has improved yet.
This is extremely good news for @bot users