Claude Opus 5 is benchmarking above Fable 5 on a lot of coding tasks at half the cost per token, so I ran both models through my real workflows to see if that actually holds up.
I gave them the same prompts across a bunch of experiments, from fixing bugs in a big codebase to building a landing page, running audience research, and coding a structural simulator. For every run I break down the cost, time, and tokens so you can figure out where each model fits in your own workflows.