Elon Musk’s Grok Bot will pick Claude over Grok when it’s better. One-model loyalty is dead.
The New Stack Matthew Burns ● Covered by 2 sources
Grok Bot will use Claude or other models when they fit the job better. Musk just admitted the cheapest loyalty is no loyalty at all.
Based on reporting by The New Stack, Matthew Burns — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Elon Musk has quietly blessed the end of one-model purity. In a Wednesday post, he said Grok Bot will use “the best back end model for any given task,” naming Claude Opus 5.5, MidJourney, Suno and other APIs. If another model gives the better result, SpaceX will use it. That’s the whole philosophy now: pick the tool that works, not the one with the right logo on it.
That admission matters because it lines up with what a lot of companies are already doing behind the scenes. At an Insight Partners event this week, the reporter interviewed 22 portfolio-company leaders, and at least 10 described the same habit: route work to the cheapest model that can handle it, save the expensive one for the tasks that really need it. One security company runs everything through a rules engine first, then hands the leftovers to smaller models. Another company uses a cheap model to figure out what the user wants before the pricier model takes over.
Cost is doing most of the pushing. One executive said the company had started with unlimited AI budgets and was now asking what all those tokens had bought. Another said staff kept choosing the most advanced model even when it was overkill, so the company started teaching people which model fits which job. And there’s a reason managers keep looking at the bill: one executive said that running a petabyte of data through any model, even a small one, would cost millions of dollars.
The switch, though, is not painless. One engineering leader told the reporter his team swapped in a newer model three days before an important demo, the workflow broke, and they rolled back over a weekend. That’s why evals matter so much: test the model before you trust it. The vendor side has its own cautionary tale too. The New Stack’s Amanda Caswell reported last month that Anthropic made Opus 5.5 20% cheaper than Opus 5 and broke four things agents depend on.
OpenRouter’s usage data shows how this plays out at scale. In the week through October 7, four of its ten most-used models were cheap “Flash” versions. Anthropic’s own cheaper tier got even cheaper on Wednesday when Haiku 5.5 launched at $.10 per million input tokens, a 90% cut for requests under 100,000 tokens. Meanwhile Claude Opus 5.5 was ninth by volume, but grew 74% in a week. The expensive model is still busy. It just isn’t getting special treatment anymore.
My take — AI-written commentary, not fact-checked reporting
This is the right call, and it should make every model vendor nervous. The era of pretending one model should do everything was always a sales pitch wearing a hoodie. The real winners are the teams that route work sensibly and stop paying flagship prices for stamp collecting.
Read more about this at: The New Stack
