On 9 June 2026, Anthropic released Claude Fable 5. They are calling it the most capable model they have ever made available to everyone, and the benchmarks back that up. It leads on almost every test they ran: software engineering, knowledge work, reading images, scientific research.
But the headline most businesses should care about is buried lower in the announcement. Fable 5 costs less than half what the previous top model cost. More power, less money. That combination is what actually changes the work we do for clients, so let me walk through what is real here and what it means if you are paying for software.
What actually launched
Two models, same engine underneath, different settings.
Fable 5 is the one everyone can use. It has safety filters switched on. Mythos 5 is the same model with some of those filters removed, and it is locked to a small group of vetted cyber defenders working with the US government, with biology researchers joining soon. Unless you are in one of those programs, Fable 5 is the one you will touch.
Anthropic explains the naming in a footnote: "Fable" comes from the Latin word for a story told, close to the Greek "mythos." The guardrails are the only thing separating them, and that is why they get different names.
The pricing is $10 per million input tokens and $50 per million output tokens. A token is about three quarters of a word, so a million tokens is a hefty document or a long working session. The point to hold onto: this is under half the price of the model it replaces, Mythos Preview.
The number that matters: months of work in a day
Benchmarks are easy to wave away. Real client work is not. So here is the example from the announcement that stopped me.
Stripe tested Fable 5 on a 50-million-line Ruby codebase. They asked it to do a migration across the whole thing, the kind of job that touches code everywhere and usually drags on. By hand, a team would have spent over two months on it. Fable 5 did it in a day.
I have run enough migration projects to know that estimate is not a fluke. The slow part of a big migration is not the clever bits. It is the thousands of small, repetitive, careful changes that a person has to make one file at a time without breaking anything. That is exactly the kind of work a model like this eats through.
For your business, this is the practical shift. Work that used to be quoted in weeks because of sheer volume, not difficulty, gets cheaper and faster. Legacy system cleanups, framework upgrades, moving off an old platform. The stuff that always felt too expensive to bother with starts to make sense. We wrote about when a rebuild is actually worth it in Custom software development: when you actually need it, and a model this strong moves that line.
Where else it got better
Coding is the loudest win, but it is not the only one.
On finance and analysis work, Fable 5 posted the highest score any model has hit on a senior-level reasoning benchmark from Hebbia, with real gains in reading charts and tables and working through documents. A trading firm, IMC, said it passed their analysis tests almost across the board.
On vision, it is now the best model for anything involving images. It can pull exact numbers off a detailed scientific chart, and it rebuilt a working web app's source code from screenshots alone. It even finished the game Pokémon FireRed using nothing but raw screenshots, where older Claude models needed a pile of helper tools to manage it.
On memory, it holds focus across millions of words in a long task and improves its own work using notes it keeps as it goes. That matters for any job that runs longer than a single prompt, which is most real work.
The pattern across all of these: the longer and messier the task, the bigger Fable 5's lead. If your problem is a quick one-liner, you will not notice much. If it is a sprawling, multi-step job, this is where the gap shows.
The catch: safeguards that sometimes get in the way
Here is the honest part, and Anthropic is honest about it too.
A model this capable is dangerous in the wrong hands. The same skill that helps a security team find a hole in their own software helps an attacker find one in yours. So Anthropic shipped Fable 5 with separate filters watching for three things: cybersecurity, biology and chemistry, and attempts to copy the model itself.
When a filter trips, your request does not get refused outright. It quietly gets answered by Opus 4.8 instead, their next-best model, and you are told it happened. Opus 4.8 is no slouch, so this is a softer fallback than a flat no.
The trade-off: they tuned these filters to be cautious, which means they sometimes catch harmless questions. Anthropic says it happens in under 5% of sessions, and they are working to loosen the filters over time. If your work touches security testing or life sciences, expect the occasional handoff, and know it is the safety net doing its job, not a bug. We covered why this kind of caution matters for businesses in Zero trust architecture explained for non-technical business owners.
There is also a data change worth knowing if you are an Anthropic business customer. Traffic on Fable 5 and Mythos 5 is kept for 30 days so Anthropic can spot new attacks and reduce false alarms. They say they will not train models on it or use it for anything outside safety, they log every human who touches it, and they delete it after 30 days in almost all cases.
How to get it, and the rollout you should plan around
Fable 5 is available everywhere now, but the access rules differ by plan.
On the Claude API and consumption-based Enterprise plans, it is fully on from day one. Developers call it as claude-fable-5.
On subscription plans (Pro, Max, Team, and seat-based Enterprise), the rollout is staged because Anthropic expects demand to outrun capacity:
- From 9 June through 22 June, Fable 5 was included at no extra cost.
- From 23 June, it dropped off those plans and needs usage credits to keep using.
- After that, once there is enough capacity, they aim to fold it back into subscriptions as standard.
If you are budgeting around this, the takeaway is simple. The free window was short. For steady use right now, plan for API or credit costs, and watch for the announcement when it becomes standard again. This is the second top-tier model Anthropic has shipped in quick succession, after the Opus 4.7 release we broke down in Claude Opus 4.7 and Anthropic's AI design tool launch, and the pace is not slowing.
What this means for the work we do for you
Strip away the benchmarks and here is the practical version.
Big, repetitive engineering jobs just got faster and cheaper. Migrations, upgrades, and cleanups that we used to quote carefully because of their size are now more affordable, and that is good news if you have been sitting on an aging system.
Analysis and document-heavy work gets a real lift, so reporting, financial modelling, and anything that means reading a lot to find the signal moves quicker.
And the work still needs a human steering it. A model that can do months of work in a day can also do months of wrong work in a day if nobody is checking the direction. The value is not the model. It is knowing what to point it at, catching where it drifts, and owning the result. That is the part we do, and a sharper tool makes it sharper, not unnecessary. If you want the plain-English version of how AI agents fit into real business work, we covered it in AI agents vs chatbots: why 2026 is the year of autonomous AI.
What to do now
If you have an aging system or a migration you have been putting off because of cost, this is the moment to get it priced again. The maths changed. Send us the system and what you want it to become, and we will tell you straight whether a model like Fable 5 makes the job worth doing now.
You can reach the team through dsrpt.com.au/services, or read more of how we think about building in the dsrpt think tank.