I was about an hour from finishing my website redesign (hope you like it!) when I hit my subscription limit. I could have stopped there, waited for it to reset, and picked back up later. Instead, I kept going, knowing full well that everything from that point on was going to cost me actual money on top of what I already pay every month. By the time I was done, it had cost me sixty dollars.

That wasn’t the smart choice, but the fact that I made it anyway says something.

Anthropic, the company behind the Claude models I’ve been using for a while now, recently released their newest and most capable model. It’s called Fable. For those of us with a paid subscription, they let you try it out using your existing plan for a short window, a few days, before it switches over to a separate, pay-as-you-go cost. That in itself was new. I’ve always had limits on what I can do with my subscription, and I’ve made peace with that. Twenty dollars a month is well worth it to me, and I’m happy to work inside those boundaries or occasionally pay a little extra when I go over, but this was the first time the newest, most powerful model wasn’t just rationed within my plan. It was carved out entirely, something I’d have to pay for separately no matter how much room I had left in my regular subscription.

I wanted to see what the fuss was about, particularly since the government had closed access to it shortly after launch because it was deemed too powerful, so I put Fable to work on two real projects. One was the third resource in my Carrying Culture project, a set of sites for teaching about Cambodians and their struggle to preserve traditional arts. The other was a long-overdue upgrade to the Hugo site you’re probably reading this on right now. (Hugo, if you haven’t run into it, is the software that turns my writing into the pages you see on the web.)

Both projects taught me the same thing: Fable is genuinely, noticeably better.

The clearest difference was how well it understood what I actually wanted. I’d describe a vision or ask for advice, and it landed on the same page with me far more often than any model I’ve used before. Less clarifying back-and-forth. Better instincts about what I meant, not just what I said. It felt more natural to work with.

It was also far more capable of running with a task on its own. At one point, while working on my site, it split itself into what are called subagents, essentially separate AI instances, each handling a different piece of the work at the same time, then reporting back so it could keep moving. It checked its own work as it went, opening up a web browser to test things and confirm they actually functioned the way they should. That last part still catches me off guard every time I see it happen. Watching an AI click around a web page to verify its own work is a strange thing to get used to. The result was a site with search (finally!) and a handful of other features I’d struggled to get working for a long time, done in a fraction of the time I expected.

What it isn’t better at, at least not yet, is writing. It’s good. It’s serviceable. But it doesn’t read the way I write, and I have a voice and point of view that guide the choices I make on the page. There’s still a gap between people who move through the world and interact, versus a model trained to predict likely next words. Fable narrowed that gap. It didn’t close it.

One more thing I noticed: left unsupervised, it was ambitious. It wanted to do more than I asked, more than I necessarily wanted, and I had to put guardrails in place to keep it from running off and generating costs I hadn’t signed up for. Which brings me back to that sixty dollars.

I hit my limit in the middle of the project, at a moment when I had the time carved out and my head was fully in it. I could have waited. I chose to keep going, watching the cost climb, because I’d already started and I liked what I was seeing and I didn’t want to lose the thread. That’s not a rational decision by any normal accounting, but it is a very human one.

I don’t think I’m unusual in this. Anthropic, and every company racing to build the next frontier model, knows exactly how compelling “the latest and greatest” is, and how willing people are to pay for that feeling even when the older, cheaper option would have done the job just fine. Fable’s quality isn’t the issue here. The hard part is making the sober, sensible choice when the flashier option is right there, working beautifully, and asking you to keep going.

Alongside Fable’s release, Anthropic also quietly changed how usage is measured for Sonnet, the model I actually rely on for most of my work day to day. The price per token stayed the same. But the way that usage gets metered changed enough that, by the numbers I’ve seen, the same work now consumes something like 40 percent more of your allotment than it used to. The sticker price didn’t move. What changed is how much of it you use up doing the exact same things you were already doing.

I don’t think this is some grand conspiracy, and I want to be careful not to overstate it. The AI industry is still competitive. OpenAI and Google haven’t raised prices on their newest models the way this metering change effectively raises Anthropic’s, and that competition is doing real work to keep things in check, but I’ve watched this shape play out elsewhere already. My streaming services all show ads now unless I pay considerably more to remove them, so much more that I don’t bother. It’s just what companies do when they need more revenue and have the leverage to take it.

What makes this particular version sting more for me is what these tools are built on. The data these models learned from came from decades of human writing, conversation, and creativity, collected without asking anyone’s permission. I’ve written about that before and I still think about it constantly. Tools built on something taken from all of us shouldn’t then become something only some of us can afford to use at their full capability. Right now, for most of what I do, Sonnet and Opus, the models included in a normal subscription, are more than good enough. But I don’t love watching the gap open up between what’s included and what costs extra, especially knowing how good competition has been at keeping that gap in check so far, and how little guarantee there is that it stays that way.

So would I tell you to go try Fable? Sure, if you’re curious and you can afford to poke at it, or if you need high-level coding power. It’s an impressive piece of work, and I don’t want that to get lost in everything else I’ve said here. But being newer and more capable doesn’t automatically make something the right tool for what you’re doing, and I don’t see myself reaching for it often. Most days, the models already included in my plan do exactly what I need.

As for the sixty dollars, I don’t regret the site. I do regret how easily I talked myself into paying it, given how clearly I could see myself doing it in the moment. I guess that knowing exactly what’s happening to you doesn’t make you immune to it.