Musk says Grok Bot will pick Claude Opus 5.5 when better

A team that runs every request through one model is on the far side of a line that moved this week. The New Stack reports that Elon Musk posted an "important note" on X about Grok Bot on Wednesday, writing that SpaceX will use "the best back end model for any given task, including Claude Opus 5.5, MidJourney, Suno and other leading APIs. Whatever is most likely to give you the best outcome."

Grok Bot is a joint product of SpaceXAI, Musk's AI lab now folded into SpaceX, and Cursor, which SpaceX agreed in June to buy for $60 billion in stock in a deal that closed in August. The New Stack frames Musk's note as "one-model loyalty is dead" and reports that at least 10 of 22 company leaders it interviewed this week at Insight Partners's ScaleUp:AI conference run their AI the same way: pick the model by the task and keep the expensive one for the work that needs it.

What triage looks like in practice

The New Stack describes two shapes from the interviews. A security company runs a rules engine over everything first and passes only what survives to small models, with larger models seeing what is left; one executive told the outlet that running a petabyte through any model would cost millions. A second company runs user requests through a cheap model to figure out what the user wants before the expensive model does anything. The reason the leaders gave was cost: one said his company started with unlimited AI budgets and is now asking what all those tokens actually bought.

OpenRouter's own ranking carries the same pattern

The New Stack cites OpenRouter's model rankings for the week through 7 October 2026, published under Creative Commons BY 4.0. Four of the ten most-used models were "Flash" models, the cheap, fast versions labs ship alongside their flagships. DeepSeek V4.1 Flash did not exist a month ago and is the top model this week. Claude Opus 5.5 is ninth and grew 74 percent in a week, the fastest in the top 10. TypeSafe's Jev decision model is tenth, and OpenAI's GPT-6 Luna Decisions is new on the trending list. OpenRouter notes that its token counts do not measure users or spend.

Why the cost gap widened on Wednesday

Anthropic launched Claude Haiku 5.5 the same day at $0.10 per million input tokens for prompts under 100,000 tokens, the price point triage setups use for the first-pass or routing model. The New Stack's Matt Burns names the pattern the ranking produces: the expensive model gets more work each week without ever reaching the top of the list.

Switching has its own risk. One engineering leader told Burns his team swapped in a newer model three days before a demo because the benchmarks said it was cheaper and just as good, the workflow broke, and the team rolled back over a weekend. His lesson was evaluations, a set of tests that would have caught the problem before his team had to find it by hand.

For a team building its own AI feature, the one-model default is no longer the free choice. The expensive model fits the hard step, the small or decision model fits the many small steps around it, and the gap is wide enough that the choice shows up in a monthly bill.

Source

Elon Musk's Grok Bot will pick Claude over Grok when it's better. One-model loyalty is dead., The New Stack, by Matt Burns. The verbatim Musk quote, the 10-of-22 conference figure, the OpenRouter ranking and the Haiku 5.5 price are all from this article. OpenRouter rankings are published under CC BY 4.0 as cited.