Fable 5 and the End of the Cheap AI Era

Claude has been the best-kept secret in the advanced AI community for a while now. While everyone else was watching OpenAI the company with the head start, the brand recognition, the household name the people actually shipping serious work quietly moved to Claude. It's been the practitioner's model: the one developers, researchers, and infrastructure people reach for when the task actually matters.

And honestly? It shouldn't be a surprise. When the Anthropic founders split from OpenAI back in 2021, they took key talent with them and, I'd argue, a lot of the spark. OpenAI kept the seat at the head of the table. Anthropic kept building. Fable 5 is what that looks like five years later.

The New Powerhouse

Fable 5 is Anthropic's first Mythos-class model made generally available, and it's state-of-the-art on nearly every benchmark that matters software engineering, knowledge work, vision, scientific research. Stripe reported it compressed a codebase migration that would have taken a team two months into a single day. It beat Pokémon FireRed with nothing but raw screenshots no maps, no helper harness, just vision.

But here's my take on what actually makes it different: it's not just raw intelligence, it's endurance. Fable 5 stays focused across millions of tokens without losing the plot. The longer and more complex the task, the bigger its lead over everything else. Previous models could sprint; this one can run a marathon and remember every mile of it. It reads its own notes, improves its own outputs, and keeps context that would have collapsed lesser models.

That's the real story. A model that can hold more and compute more without drifting is worth more in practice than a few benchmark points.

Fable vs. Mythos

Quick clarification, because the naming confuses people: Fable 5 and Mythos 5 are the same underlying model. Mythos 5 is the unrestricted version currently limited to a small group. It has the strongest cybersecurity capabilities of any model in the world.

Fable 5 is the version made for the rest of us. Anthropic added what I can only describe as parent mode: safeguards that intercept queries on sensitive topics and quietly hand them off to Claude Opus 4.8 instead. Anthropic admits the filters are tuned conservatively they'll sometimes catch harmless requests but they trigger in less than 5% of sessions on average. A small tax for access to a Mythos-class model.

The Other Signal: Cheap AI Is Ending

Here's the part nobody wants to hear. Fable 5 launched included on paid Claude plans, got yanked offline June 12 by a US export-control directive, came back July 1 and now the clock is ticking again. Through July 7, Fable 5 is included on Pro, Max, Team, and select Enterprise plans for up to 50% of your weekly usage limits. After July 7, it moves to usage credits billed at API rates.

And those API rates are not cheap: $10 per million input tokens, $50 per million output — double the price of Opus 4.8, and expensive compared to basically any mainstream LLM on the market. For good reason. It's powerful. But powerful and affordable are officially parting ways.

To be fair, Anthropic says this is a capacity decision, not a permanent paywall they've stated they intend to bring Fable back into standard subscriptions once demand stabilizes. Maybe they will. But the direction of travel is clear: frontier capability is becoming a metered, enterprise-grade line item. The era where the best model in the world came bundled in your $20 subscription is closing.

What You Should Do About It

Two pieces of advice:

First: learn to offload work to local AI. Open-weight models are genuinely good now, and for a huge chunk of everyday tasks (summarization, drafting, routine code, classification) they're more than enough. Build the habit of routing work by tier: local and cheap models for the volume, frontier models only for the tasks that measurably need them. That discipline used to belong to API teams. It's about to belong to all of us.

Second : you have a couple days. If you're on a paid Claude plan, Fable 5 is included through July 7. Every project, every idea, every "someday" task sitting in your notes app that needs serious horsepower run it now. Refactor the codebase. Draft the whole spec. Feed it the 400-page document. This is your last chance to do it or better said, your last chance to do it cheap.

After July 7, the meter starts running.