Anthropic Just Shipped Its Best Coding Model to the Public — Free for 13 Days, Then It's Metered. Here's How to Spend the Window.
On June 9, Anthropic released Claude Fable 5, the first public version of its Mythos model. The headline number: 80.3% on SWE-Bench Pro, about 11 points above Claude Opus 4.8 and more than 20 points ahead of GPT-5.5. It's the strongest coding model you can put your hands on right now. And through June 22, it's included in Pro, Max, and Team plans at no extra charge.
On June 23, that ends. Fable 5 gets pulled from those plans and moves to usage credits: $10 per million input tokens, $50 per million output. That's roughly double Opus 4.8's pricing. So the real story here isn't the benchmark. It's the 13-day free window, and what a solo operator should actually do with it.
The free window is a funnel, not a gift
Anthropic is not giving you two weeks of frontier inference because it's feeling generous. It's giving you two weeks to wire Fable 5 into your workflow, watch it clear tickets your previous model stalled on, and develop the muscle memory of reaching for it. Then the meter turns on, and the switching cost it just built keeps you paying.
That timing is a feature, not a bug. And the right response isn't cynicism. It's to use the window on purpose instead of drifting into a metered habit you didn't decide to have.
Here's the mental model I'm using. For the next two weeks, Fable 5 is free compute that's better than what I normally pay for. The question is which work is worth front-loading into that window, and which work I'd never pay $50-per-million-output to run once the window closes.
What to route to Fable 5 before June 23
Front-load the hard, bounded, high-impact stuff, the work where an 11-point jump on a real software-engineering benchmark actually changes the outcome:
- The refactor you've been avoiding because your current model keeps half-finishing it. Big, gnarly, multi-file changes are exactly where the SWE-Bench Pro gap shows up.
- A test suite for the part of your codebase that has none. Generating coverage is tedious and bounded, perfect for a model that's free this week.
- The migration or upgrade you've been putting off. If you've got an Astro or framework bump sitting in a branch, this is the week to let the better model grind through it.
- Anything where you'd normally babysit a weaker model through three rounds of corrections. The fewer steering loops, the more the quality difference compounds.
Do this work now, while the marginal cost is zero, and bank the output. A refactor you ship this week is a refactor you don't pay frontier rates for next week.
What stays on Opus 4.8 after the window closes
Once the meter's on, the math changes hard. At $10/$50 per million, Fable 5 is a precision instrument, not a daily driver. The work that should stay on Opus 4.8, or a cheaper model entirely, is everything where the quality ceiling doesn't matter:
Boilerplate, simple CRUD, doc generation, commit messages, the hundred small completions a day that any competent model handles fine. Paying double for those is lighting money on fire to feel like you're on the best model. Opus 4.8 is still excellent, and for most of what a solo builder does in a day, the difference between it and Fable 5 is invisible.
The decision Fable 5 forces is the same one every tiered model lineup forces: route by difficulty, not by reflex. The free window is your chance to learn, at zero cost, exactly which of your tasks genuinely need the top model, so that on June 23 you're paying for the 10% that benefits and not the 90% that doesn't.
The Mythos 5 split tells you where the moat is
There's a second detail worth sitting with. Fable 5 is the public model. It blocks responses in high-risk domains (cybersecurity, biology, chemistry) and falls back to Opus 4.8 when you wander into them. The actually-unrestricted version, Mythos 5, is gated to roughly 150 organizations working on critical infrastructure and cyber defense.
So the frontier now has two tiers: the one you can buy, and the one you have to be approved for. For a solo operator, that's not a complaint. It's a signal. The capabilities being held back are exactly the ones with the highest liability and the strongest demand. If your work lives in the safe, public band of model capability, you're competing in the part of the market that gets commoditized fastest. The defensible work is increasingly the work that sits next to the gated capabilities, not on top of the public ones.
What I'd actually do
Treat the next 13 days as a free sprint, not a free trial. Pick the two or three pieces of hard, bounded work you've been deferring and run them through Fable 5 this week while it costs nothing. Ship them. Then, before June 23, write down the three categories of task where the better model actually changed your output, and wire only those to the credit-metered version. Everything else goes back to Opus 4.8.
The trap is the one Anthropic is counting on: letting "the best model" quietly become your default for everything, then absorbing a credit bill in July that you never decided to sign up for. The window is genuinely useful. Just spend it like it's going to close, because it is.
Author
Lukas
@lukcombinator