I like to redesign this blog every once in a while, and this time it lined up with Anthropic releasing Fable 5, their new model that sits above Opus. The redesign seemed like a good test for it: it’s a real codebase, I have opinions about how the site should look, and if the result was bad I could just revert.

I’ve tried this with earlier models and I always had to wrangle with them to get the result looking decent enough. This time I didn’t, so here is what it did well, and why the way they’re offering the model looks like a stunt to me.

The Redesign⌗

The old theme was an off-the-shelf Hugo theme that I had been adjusting to my needs: a card grid on the homepage, a mobile dropdown menu held together with JavaScript. It worked, but I was getting bored of it.

The inspiration this time came from mitchellh.com, so the prompt was along those lines: editorial look, monochrome, one accent color. Vague on purpose, I wanted to see what it would do with room to make its own decisions.

It came back with a sticky sidebar, an “m.” mark, a red dot marking the active nav item, a single reading column, and the homepage as a plain list of every post with hairline separators. Signal red is the only color on the site. Four commits total:

75 files changed, 1728 insertions(+), 1710 deletions(-)

Roughly the same size it found the theme at, just rewritten.

What Made Me Keep It⌗

The visuals are one thing, but the reason I merged this is the structure underneath. It split the SCSS into layers: design tokens at the bottom (colors, type scale, spacing), base styles on top of that, then one file per component. Every themable value is a CSS custom property, so dark and light mode are now two blocks of variables instead of overrides scattered across the whole theme.

I would not have done this myself. I would have changed the colors on top of the old mess and called it a day (which is more or less what my previous redesigns were).

It even handled the asset work. The favicon is new, an “m.” mark with a red period matching the accent color, and Fable made it itself, all the icon sizes included.

The Pricing⌗

Now the catch. Fable costs $10 per million input tokens and $50 per million output, double Opus 4.8. It’s included on the regular Claude subscriptions, but only until June 22, that’s straight from the announcement. After that it moves to usage credits.

The Hacker News thread on the launch shows both sides of the trade. People describe it finding bugs that earlier models had introduced and writing a fuzzer unprompted to verify its own work, and right next to that, people hitting their usage limits in half an hour where the same plan used to give them five or six hours of agent work.

So they’re not even hiding the playbook: give everyone the best model for two weeks, let people build their workflows around it, then move it behind pricing that pays for the GPUs. The official line is that this is about capacity and that they intend to bring it back to subscriptions as soon as they can. I doubt it, there’s no money in serving the most expensive model on a flat monthly fee. Once people are hooked, usage credits make a lot more sense. I wrote recently about not being able to hear an AI announcement without asking who profits from it, and I can’t turn that off here either.

But if I follow that logic through, the stunt only works if the model is worth getting hooked on, nobody builds a workflow around a mediocre model just because it’s temporarily cheap. Going by what it did to this blog, it might be. Will I keep using it? Not sure, it spends a lot of tokens and I’m not convinced it’s necessary for most of the things I do with AI. For the occasional bigger job like this redesign, maybe, while it’s still on the subscription.

All in all I’m glad I tried it out, and I do prefer how the blog looks now compared to before. But no one should lie to themselves thinking these companies want to give out additional benefits on their own. I’m seeing a lot more content about companies starting to look at their AI costs, so if providers keep raising prices and locking new models behind additional fees, we might have to rethink some things that became the norm in the past few years.