Musk Says Grok Bot Will Run on Anthropic's Claude Opus 5.5 — 'The Best Back End Model for Any Given Task' — With Midjourney and Suno in the Mix

Elon Musk said on Tuesday that xAI's Grok Bot will stop being a Grok-only product. "Going forward, SpaceX will use the best back end model for any given task, including Claude Opus 5.5, MidJourney, Suno and other leading APIs," Musk wrote in a post on X. "Whatever is most likely to give you the best outcome." The post effectively turns xAI's persistent agent into a model-agnostic orchestration layer — one that names its chief rival's flagship model first.

Grok Bot, which entered public beta on August 11, is xAI's always-on agent: each bot runs on its own cloud computer with a browser, filesystem and terminal, can sign into applications, carry out multi-step work and keep going after the user closes their laptop. Until now the assumption — encouraged by xAI's own marketing that Grok 4.7 was "trained to natively understand the Grok Bot harness" — was that the bots ran on Grok models.

Within hours, an xAI engineer working on Grok Bot filled in the most important detail. "All bots will be powered by Opus 5.5, and they can spawn cloud agents of whatever Cursor models we have," wrote Lauren (@poteto) on X. That makes Claude Opus 5.5 the default brain of every bot rather than an occasional fallback, with Cursor-hosted models available as spawnable cloud agents. xAI's own launch table for Grok 4.7 explains why: the model wins on legal and electrical-engineering tasks, but scores 37.6% on multi-hour terminal work against 57.9% for Anthropic's Fable 5.1.

The commercial geometry is unusual. xAI already rents its Colossus supercomputer — more than 220,000 NVIDIA GPUs by earlier reporting — to Anthropic, which uses the capacity to serve Claude Pro and Max subscribers. Routing Grok Bot tasks to Claude makes SpaceXAI an Anthropic customer as well as its landlord, and gives Anthropic a fresh distribution channel just as the lab prepares for a widely expected IPO. Musk framed the change as outcome-first: improving Grok itself is no longer the only way to improve the bot.

What Musk did not say is nearly as important as what he did. There is no rollout date, no explanation of how tasks get assigned to which provider, no pricing (Opus 5.5 is among the most expensive models on the market), and no word on whether users can see or choose which model handled a step. Data-handling questions are also open: a bot holding CRM records or invoices that hands one step to Claude, Midjourney or Suno is sending context to a new recipient, and neither Musk's post nor xAI's documentation specifies filtering, per-provider controls or logging.

The strategic read is bigger than one product. If the persistent agent environment — memory, files, tools, approval rules — is the product, the foundation model underneath becomes a swappable supplier, and the orchestrator that consistently produces the best result wins the user. That is a bet Anthropic, Midjourney and Suno are happy to monetize, and an uncomfortable one for every lab that assumed owning the best model would be enough.