The headline about Grok Bot 1.0 is shared teammates. The question almost nobody asks is which model is actually doing the work.
xAI's product page does not answer it. The documentation describes behaviour, a Bot with its own cloud computer that signs in to tools and works through tasks on a schedule, but never lists the models behind it or says whether different tasks route to different ones. A 9to5Mac timeline in October added a detail: the Bots had begun using Anthropic's Claude.
xAI has not confirmed the current model mix. Musk was more direct in October, saying the agent would also draw on models beyond Grok, naming Claude Opus 5.5, MidJourney and Suno.
On engineering grounds this is ordinary. A Bot that operates a desktop is mostly an executor, and a sensible executor picks the right tool: a coding model for code, an image model for visuals, an audio model for music, a cheap small model for trivial steps. Multi-model routing is standard for this generation of agent products, not a scandal.
Commercially, the implications are different.
First, data terms. An enterprise buyer signing a contract needs to know where data goes. If part of a task is processed by another lab's model, data-processing agreements, regional retention and audit scopes all have to be recalculated. Second, margin. Upstream price changes pass straight through to a $30 seat that xAI does not control. Third, competition. An agent running partly on a rival's models is structurally late: the rival ships its new version to its own products first, and your users' experience follows someone else's roadmap.
The structure behind it
xAI closed its $60 billion acquisition of Cursor's parent Anysphere on August 14. Grok Bot's security page references Cursor single sign-on and privacy mode, and the beta was bundled with Cursor plans. Cursor is itself a multi-model product, which goes some way to explaining why Bot never intended to run on Grok alone.
For buyers the test is simple: can you switch. If the model is an opaque box, a customer whose quality drops has no lever except waiting. If the product lets you pick a model per task, or at minimum publishes a model list with a change log, procurement can put it in a critical path.
Right now xAI offers the former. The product page does not discuss models, the launch post does not discuss models, and the only signals are third-party reporting and the founder's posts. For a product whose selling point just became team colleague, that is the single easiest thing for an enterprise security review to block on.
Three things to watch: whether xAI publishes a model list; whether enterprise plans allow naming or excluding providers; and whether Claude disappears from the chain once Grok's own next version catches up. The answer to the last one will say more about the product's real position than any benchmark.
xAI Won't Say Which Models Run Grok Bot. Reports Say Some Tasks Already Use Claude
← Previous: Grok Bot 1.0 Turns a Personal Agent Into a Shared Teammate in Slack, Teams and Calls
Next: →Grok Bot Gets the Keys to Your Mac: Messages, Mail, Notes and Four More →