Using Up Fable 5: The Right Way to Test a Top-Tier Model Is Your List of Past Failures

I used up my subscription allowance for Fable 5 cleanly, the night before it expired. Honest summary: had I not paid for access at this particular moment, I would be uneasy about my own work going forward. Here is what made it worth that.

The turning point was retrying problems I had given up on

Early on, I did not really see what the fuss was about. The models I already had were capable enough.

What changed my view was throwing it a problem I had never been able to solve, on a whim, and watching it come apart easily. From then on I switched to working through every project I had written off as impossible, and results kept coming. The right way to test a top-tier model is not your daily work but your list of past failures.

The biggest gain: breaking the everything-looks-the-same problem

The result that mattered most to me was solving a long-running frustration with web design. Have AI design something and it comes out clean but reminiscent of something you have seen before. Previous models were less a collaborator than a supplier of templates.

Fable 5 — with GPT-5.6 Sol, below — built me the system that solves it. I will keep the details to myself. But the point is the systematizing: once a capable model has built the structure, you can run it on other models afterward.

GPT-5.6 Sol took over, but running both is better

GPT-5.6 Sol arrived just before the Fable 5 window closed and has been more than adequate as a successor. The ideal, though, is running both. Design in Claude and review in Codex; have Claude check what Codex implemented. Setting two vendors’ models to find fault with each other’s work has a clear effect on quality.

Cost is equally real. Maintaining top-tier plans on both is a serious expense, and which one to keep is a personal call. In my case, staying anchored to OpenAI, which I have used longest, and paying for Claude only during the periods I need it looks like the practical answer. That conclusion will change with the next model, of course.

Enjoy the golden window

An extension of the Fable 5 period, a temporary loosening of GPT-5.6 limits — competition between the two companies keeps producing generosity toward users. A great deal of work moved as a result, systems were left behind, and my external brain of accumulated notes is in better shape. While the competition lasts, use it hard.

This article is adapted for AI Jiten from testing notes originally published on the author’s blog, Delaymania.