A few days ago, Theo shared his AI model tier list. He placed Fable 5 in its own category. I agree; Fable 5 is a beast. But, you know what? Who cares!

While it’s important to know model capabilities, most people and businesses will never use the full capabilities of these models; nor do they need to. If everything stopped today and we were left with the current list of models, the capability gap would be immense.

Ask yourself this question: if Fable 5 were free, what would you have it do?

Yes, what’s missing is vision and ambition.

Anyway, it’s not about which model is best. It’s about asking: best at what, under which constraints, and at what price?

Intelligence is becoming one variable among several: reasoning quality, coding ability, latency, inference cost, context, tool use, reliability, privacy, deployment flexibility, etc. Cost and speed increasingly belong alongside raw intelligence when evaluating models.

We’re moving from model selection to model portfolios.

A company might have a frontier reasoning model for difficult analysis, a coding-specialized model for engineering, a cheap fast model for classification/extraction, an open model for high-volume internal workloads, and perhaps specialized models for voice, vision, or other functions.

The model wars are becoming less strategically important at the application layer. Not because models don’t matter. They matter enormously. But because the market is producing many forms and prices of intelligence.

The scarce capability therefore shifts upward: Knowing how to apply intelligence.

Models become inputs. Workflow design, orchestration, context, data, evaluation, judgment, and organizational redesign determine whether those inputs create economic value.

That’s also why I think a simple model tier list is becoming less useful even as a fun exercise. It’s not about which model wins. It’s which combination of models produces the best system?

That’s a much more interesting question, and much closer to how businesses should actually think about AI.