Found in Fable 5’s Shadow: Opus 4.8 Is More Than Enough for Everyday Work
Fable 5 took all the attention, but the other thing I took away from a short, intense period of use was this: Opus 4.8 is more than enough for everyday work.
The jump to a top-tier plan
Until now I had used Codex on a standard paid ChatGPT plan and honestly found that sufficient. To try Fable 5 I subscribed to Claude’s top tier for the first time.
The result exceeded what I expected. The jump from the standard tier to top-tier models is felt more than the capability numbers suggest. Someone who had balked at the idea of paying that much monthly for AI ended up reconsidering it as an investment that might be necessary at this point in my work.
Design with Fable 5, implement with Opus 4.8
Most of the period ran as a division of labor: design and planning to Fable 5, implementation to Opus 4.8. What surprised me was how capable Opus 4.8 turned out to be. It writes accurate code, and it holds up against Codex, which I know well. I reached the point of thinking I would keep the plan just to keep using it.
Fable 5 is genuinely in a different class, but for daily work it is frequently overspecified. Fable 5 for the hard problems, Opus 4.8 as the everyday partner — that split is where cost and capability settle comfortably.
Hard to separate what was actually impressive
Looking back, the awkward part is that I cannot cleanly separate what impressed me: Fable 5 itself, Claude Code as a tool, or simply the headroom of a top-tier plan. Someone who already used Claude Code daily would have come away with a different impression.
The one thing I can say with confidence is that when something new appears, you should get your hands on it. Reading comparison articles without touching it would not have conveyed this difference.
Where it left me
A backlog of work started moving in a good direction, and problems I had left unsolved got resolved. Joining the excitement around a top model while quietly finding the everyday answer behind it — that is another way to get value out of a trial window.
This article is adapted for AI Jiten from testing notes originally published on the author’s blog, Delaymania.
