I pushed my usage of the world’s most powerful AI model to 90% this week — on purpose.
Because today, it leaves my subscription.
For a short window, Fable 5 (the most capable model available) was included in the Claude plan. As of today it moves to usage-based pricing, because demand is outpacing what Anthropic can serve at a flat rate. (Temporary, they say.)
So I front-loaded. I ran everything meaningful through the best model on earth while access was effectively free.
Here’s the lesson for anyone building with AI.
You have to stay on top of these models — and move to the newest one the moment you can.
Here’s a reason most people miss. When a new model launches, the older ones tend to quietly get worse.
Think about the mechanics: providers put their best and most compute behind the newest model. Compute doesn’t appear from nowhere — so it gets pulled from the older ones.
The model that felt razor-sharp six months ago isn’t running on the same firepower today. So you have to adapt.
Most engineers do the opposite of what they should. They ration themselves on a cheaper, weaker model to save a few dollars, while the best model in the world sits one click away.
Comfort is the trap. The moment you settle on a model, the ground is already shifting under it.
When a new model drops, how fast do you actually move your work onto it?
Stay at the frontier,
Kirill
P.S. I maxed out Fable while I could. But the real edge isn’t one lucky window — it’s never standing still while the model under you quietly loses horsepower.