There’s a growing narrative in the AI space that goes something like this: just use the most powerful model available. Why settle for second best? The extra cost pays for itself.
It sounds logical. And sometimes it’s true. But most of the time, it’s terrible advice.
The Real Cost of “Best”
This week, Anthropic published a blog post about Claude Code modernizing COBOL — a legacy programming language that powers the backbone of global finance. IBM’s stock dropped 13% in a single day. $40 billion in market cap, gone. From a blog post.
The gap between good AI and great AI is real. No one’s arguing that.
But here’s what the “always use the best model” crowd gets wrong: the most expensive model isn’t always the best model for the task. And using it for everything isn’t a power move — it’s waste disguised as sophistication.
When Good Enough Is More Than Enough
Sonnet 4.6 is now benchmarking neck and neck with Opus on most real-world tasks — at roughly one-fifth the cost. For the vast majority of daily workflows, it’s not just adequate. It’s excellent.
Writing emails. Summarizing documents. Generating scripts. Cleaning up data. Drafting marketing copy. Answering straightforward questions. Sonnet handles all of this without breaking a sweat. Running Opus for these tasks is like hiring a neurosurgeon to put on a Band-Aid.
When You Actually Need the Big Gun
Then there are the moments where Opus genuinely earns its keep.
You’re debugging a gnarly issue across a massive codebase and the cheaper model keeps going in circles. You’re making architecture decisions where one wrong call means three hours of rework. You’re doing high-stakes financial or legal analysis where missing a single nuance costs real money. You’re navigating a sensitive negotiation where tone and subtlety actually matter.
These are brain surgery moments. And yes, you want the best surgeon in the room.
The Real Flex
The companies getting the most out of AI right now aren’t the ones defaulting to the priciest model for everything. They’re the ones building systems that route the right task to the right model.
Simple query? Haiku. Standard workflow? Sonnet. Complex reasoning that demands precision? Opus.
That’s not cutting corners. That’s intelligence about intelligence.
The AI cost conversation is shifting. It’s no longer about whether you can afford the best model. It’s about whether you’re smart enough to know when you actually need it.
Because the most expensive mistake in AI isn’t using the wrong model. It’s using the right model for the wrong reason.
Here’s the CNBC article covering it: https://www.cnbc.com/2026/02/23/ibm-is-the-latest-ai-casualty-shares-are-tanking-on-anthropic-cobol-threat.html
Want the Bloomberg one too? https://www.bloomberg.com/news/articles/2026-02-23/ibm-shares-plunge-as-anthropic-touts-cobol-modernization-efforts