I shipped 3 Claude Skills the week before Fable 5 launched. Here's what changes now.

Here's the whole thing in three sentences. Anthropic released Claude Fable 5 on June 9, the first Mythos-class model the public can actually use, and it's free on Pro and Max until June 22. I'd spent the week before its launch shipping three small things: a scope-cutting skill called enough, a Shopify Theme Audit Skill, and a macOS app called Alter that puts AI inside your clipboard. Below is what changes now, what doesn't, and what I'd build next if you handed me a Pro plan and ten days.

What Fable 5 actually is, in plain language

Anthropic ships two model families. Mythos is the internal frontier model the company has been running for safety research. Fable is the public-facing version of that same model, with the sharp edges sanded down. Fable 5, shipped on June 9, is the first time a Mythos-class model has been made generally available to anyone who can pay.

The pricing is ten dollars per million input tokens and fifty dollars per million output tokens. That's double Opus 4.8. It's available on the Claude API, AWS Bedrock, Vertex AI, and Microsoft Foundry. On Pro, Max, Team, and seat Enterprise plans it's free from June 9 through June 22, after which using it on those plans burns usage credits.

The headline result is software engineering. Anthropic's own framing is that Fable 5's lead over previous Claude models grows as tasks get longer and more complex. Translation for the rest of us: it doesn't lose the thread on multi-hour work. That is the single thing that mattered most to me as a builder, and the rest of this post is about why.

The June 22 thing matters more than the launch

Anthropic gave the world thirteen free days on a model that costs fifty dollars per million output tokens. That's not generosity. That's a controlled test of how the world uses Mythos-class capability before they charge for it. If you have any interesting project that's been waiting for a smarter model, this is the window. After June 22, the math changes hard for anyone who isn't already paying for tokens.

I'm using my ten days the way I'd use a free pass to a power tool I couldn't afford. I'm pointing Fable 5 at the longest, ugliest work in my backlog. Not chat. Not edits. The real grind.

What I shipped the week before, and why I didn't wait

I get asked a lot why I keep shipping things instead of waiting for the next model. The honest answer is in my operator-mode post, but the short version is that waiting is how you end up with a year of unshipped ideas. So in the week leading up to Fable 5, I shipped three.

Enough, a Claude Skill that asks one question: when do you stop building. It runs in three modes. Pre-build for testing whether the idea is worth starting. Mid-build for finding what to cut. Post-launch for deciding which feature request is a real signal. It uses eighteen cut-test patterns. No accounts, no upsells, one file long. I built it because I was getting tired of debating scope with myself.

Theme Audit Skill, a Shopify-specific Claude Skill that audits any theme in five minutes. Twenty-two automated checks across performance, accessibility, Liquid quality, and conversion. It returns a 0 to 100 score, severity-tagged findings, exact file and line numbers, and copy-paste fixes. I built it because I'd audited the same fifteen things on enough D2C themes that it deserved to be automation, not me.

Alter, a macOS app that puts AI inside your clipboard. One hotkey, any app, sub-two-second response. I built it because I'd counted my own ChatGPT tab switches and the number was forty a day. Forty switches, each one a small tax on attention. Alter kills the switch.

Three things, one week, all on Opus 4.8. None of them needed Fable 5 to ship. All of them get better with it.

What Fable 5 actually changes for a builder

I want to be careful here because most takes on a new model are vibes. Here's what I've seen in the first seventy two hours of using it, on real work, not benchmarks.

The biggest shift is sustained attention. Fable 5 can hold a much larger working context and not start hallucinating about what it did three steps ago. On Opus 4.8, my Theme Audit Skill needed to be carefully scoped to one section at a time. On Fable 5, I can hand it a whole theme and ask for the full audit in one go, and it doesn't lose track of which files it's already looked at.

The second shift is plan quality on first try. With Opus 4.8 I'd usually have to nudge the plan once before it was good. With Fable 5 the first plan is more often the plan I would have nudged it toward. That sounds small. Across a week it adds up to a meaningful number of cycles.

The third shift is the one I didn't expect. Fable 5 is noticeably better at saying no. It will tell you that a feature you asked for is a bad idea, that a refactor isn't worth it, that the simpler version of your prompt is fine. Opus 4.8 is friendlier. Fable 5 is more honest. For my enough skill, where the whole job is to push back on scope, this is a huge upgrade. The verdicts get sharper.

What it doesn't change

It doesn't make me faster at small tasks. A two-line CSS fix is a two-line CSS fix. Sonnet 4.6 is still my default for quick edits and Haiku 4.5 is still the right choice for cheap routing work. Fable 5 is overkill for ninety percent of what I do day to day, and at fifty dollars per million output tokens, overkill has a price tag.

It doesn't fix bad inputs. If your prompt is vague, Fable 5 produces a more confident vague answer. The bar on prompt clarity goes up, not down, because a smarter model is more rewarded by precision.

And it has hard limits the previous models didn't. The Register reported that Fable 5 refuses innocuous prompts more often than Opus 4.8 in certain categories. Fortune covered an accusation of "secret sabotage" where some researchers say capability is being throttled in subtle ways. I haven't seen the worst of this in my own work, but I have hit a few refusals on things I consider clearly fine. If you're doing security research or anything that pattern-matches to the model's safety filters, expect more friction than before. In those cases Anthropic says the model falls back to Opus 4.8, which is a sensible escape hatch.

Three skills I'd build next, because of Fable 5

A model is only as useful as the workflow you wire it into. Here's the next three I'd build, and I'm naming them out loud because if I don't, I'll spend the rest of the week debating them in my head instead of shipping.

A migration skill for Shopify themes that takes an old Online Store 1.0 theme and produces a 2.0-ready version, with section-by-section diffs, a checklist of redirects, and a rollback plan. This wasn't realistic on Opus 4.8 because the model would drop context halfway through a complex theme. On Fable 5 it might be a one-prompt job.

A GEO audit skill that runs the AEO and GEO checks I wrote about last week against any storefront. It would check the robots.txt for blocked AI crawlers, look for missing Product schema, run a handful of category questions through Perplexity, and tell you in plain English where you don't show up.

A scope-creep alarm that hooks into a real codebase and watches for the patterns Enough catches. Today Enough is reactive, you run it. The next version sits in your repo and pings you when a PR starts drifting outside the original scope. Fable 5 is the first model I'd trust to make that call without crying wolf every other commit.

I'll ship one of these inside the free window. Probably the GEO audit, because the timing on AEO content is right and the work overlaps with what I already do for clients.

The honest part nobody on launch day will say

A new state-of-the-art model is not the same as a new state-of-the-art product. The gap is the workflow. Fable 5 will be misused by ninety percent of the people trying it this week. They'll point it at chat. They'll ask it to write a tweet. They'll burn the free window on tasks Haiku could do for a thousandth of the cost. And on June 23, they'll go back to whatever they were doing.

The five percent who get the most out of these ten days are the ones who already have a real, ugly, multi-hour piece of work waiting for a smarter model. If that's you, point Fable 5 at it tonight. If it isn't, the more useful exercise this week is to find the ugly piece of work, not the new model. The model will still be there next month at the right price for the task that earns its keep.

What to do before June 22

Five things, in order of return on effort.

1. Open Claude on your Pro or Max account and confirm Fable 5 is showing up in the model picker. If it isn't, restart the app. If it still isn't, your account may be on a plan that doesn't include it.
2. Pick the single longest task in your backlog. Not the most interesting one. The longest one. Hand it to Fable 5 in one go, with a plain prompt, and watch how it handles step five, step ten, step twenty.
3. Run the same prompt on Opus 4.8 and compare. The point isn't to win an argument with yourself. It's to know, concretely, whether the gap on your work justifies the price after June 22.
4. If you build, ship something. A skill, a script, an internal tool. Anything that captures the workflow you discovered. The free window is a forcing function. Use it.
5. Write down what didn't work. The refusals, the dead ends, the places it was confidently wrong. That list is more useful than the wins, because it tells you where Opus 4.8 is still the right tool.

Close

Fable 5 is not the thing that changes your year. The thing that changes your year is whether you point any model at work that matters, finish it, and ship. The models will keep getting better. The pattern of shipping won't change.

If you want help figuring out which Shopify work is actually worth pointing Fable 5 at this week, that's the kind of thing I do. Send me your store and the longest thing on your backlog. I'll tell you whether it's the right test, and if it isn't, what is.