Fable has better taste, Sol has a better price, and neither deserves your loyalty

The model has been out for a day, so anything I say now comes with that caveat. But my early read on GPT-5.6 Sol is that it is almost as good as Fable. Almost. What is already clear is that Fable has better taste. Put the two side by side on a UI build and it is not close, Fable wins hands down. Sol gets you most of the way there. Fable gets you the last mile, the part where the thing actually looks and feels right.

Here is the complication: Sol is significantly cheaper. Depending on the task it runs anywhere from marginally cheaper to about a third the cost of Fable. So if you are paying per token and you care even a little about the bill, it is going to be hard to keep justifying Fable over Sol for most of what you do.

Unless you are on a subscription, which most of us are

The catch is that most of us are not paying per token. We are on subscriptions, and if you are on a $200 max plan you can still get an enormous amount of Fable out of it, at least until July 12. That is the day Fable is supposedly coming off subscription, after which it looks like we start paying usage rates that make it prohibitively expensive for most of us.

My hope is that Sol being included in OpenAI's subscription forces Anthropic's hand and keeps Fable on theirs. If that happens, I will keep using Claude for almost everything I do. If it does not, and we effectively lose access to Fable, there is a very real chance I move to ChatGPT. I do not say that lightly after two years on Claude, but the math is the math.

Pay-per-use is where it gets interesting

If you are stuck paying usage, maybe you are on a big enterprise plan, it is not even clear Sol is the right pick. Yes, Sol gives you nearly Fable performance at a third the cost. But Grok 4.5 now looks like it gives you nearly Sol performance at a third of that. And for some work, Meta's Muse Spark gets you close to Grok at about half the cost again. We just watched that whole cascade reprice in a single day.

Just when this looked like it was settling into a two-horse race, the field reopened. Grok 4.5 inside Cursor is the one blowing people away, reportedly on par with Opus at roughly a third the cost. xAI is officially back in the conversation. This is the fork I wrote about widening into something more like a four-way split.

The lesson is the same one it always is

The takeaway for any business is not which model wins this week, because the answer will change next week. It is this: do not put all your eggs in one basket, and do not get attached to any single model or platform. The attachment is the risk. The thing that was clearly best in June is the third-best value in July.

The way you protect yourself is to keep your knowledge base outside of these platforms. For our clients right now we are building new knowledge bases inside GitHub. Once your prompts, your context and your skills live in a plain repo, switching is trivial. You just point whatever model or platform is winning this month at the repo and go. Loyalty to a model is a bet that it stays ahead. Nothing in the last two days suggests anything stays ahead for long.

Opinion columns reflect the personal views of the author. Our reporting on the stories referenced here lives on the linked pages.