On June 9, Anthropic released Claude Fable 5, the smartest AI model it has ever sold. It is noticeably better at hard work, it costs twice as much to run, and it ships with new safety rails. Here is what it actually means for your business, minus the jargon.
You may have seen the headlines this week. Anthropic, the company behind the Claude AI that powers our agents, has released a new model called Claude Fable 5. The tech press is calling it a big deal, and for once they are right. It is also easy to misread what it means if you run a business rather than a software team.
The short version:
Think of Claude as a range of engines from the same maker. Until this week the line-up had three tiers: Haiku, the small fast one for simple jobs, Sonnet, the everyday workhorse, and Opus, the powerful one for complex work. Fable 5 now sits above all three. A new top tier.
The interesting part is where it came from. Anthropic has long kept a class of models internally that it considered too capable to sell, a tier it calls Mythos. Fable 5 is the first model from that class released to the public, as TechCrunch reported, days after the company publicly warned that AI is becoming dangerously capable. What made the release possible is a new set of guardrails, which we cover below.
Anthropic says Fable 5 is its biggest jump in capability yet, with the largest gains in building software, knowledge work like reports and analysis, reading images and documents, and scientific research.
Two numbers translate well into business terms.
First, the quality jump: more than 10% better than the previous top model on some benchmark tests. In AI, where progress usually arrives a percent at a time, that is a leap.
Second, the reading capacity: Fable 5 can take in about 750,000 words in a single request. That is every contract in your filing cabinet, or several years of email, read in one pass and reasoned about as a whole. It can also produce very long pieces of work in one go, a full report or a working piece of software, without losing the thread.
AI models are priced on the amount of text they read and write. Here is the current Claude line-up, and where the new model lands. Prices are in US dollars.
| Model | Best for | Reading cost | Writing cost |
|---|---|---|---|
| Fable 5 (new) | The hardest work: complex builds, deep analysis | $10 | $50 |
| Opus 4.8 | Complex everyday work, autonomous agents | $5 | $25 |
| Sonnet 4.6 | The everyday workhorse | $3 | $15 |
| Haiku 4.5 | Simple, high-volume jobs | $1 | $5 |
Reading and writing costs are per million tokens, the unit AI is billed in. A token is roughly three quarters of a word.
To make that concrete: asking Fable 5 to read a 100-page document and write a 10-page summary costs around one US dollar. The same job costs about 50 cents on Opus 4.8, and about 10 cents on Haiku. The skill, and it is the heart of what we do at TurnkeyAI, is matching each job to the cheapest engine that does it well.
This release came with a genuinely new safety setup. Fable 5 watches its own answers in real time. If someone asks about truly dangerous territory, building weapons or attacking computer systems for example, the model blocks its own response and a safer previous model answers instead, as NBC News detailed. That guardrail system is the reason Anthropic was willing to sell a Mythos-class model at all.
For your business the practical takeaway is simple. Normal work, quotes, reports, customer replies, analysis, will never bump into these rails. They exist for the edge cases you would never go near anyway.
Here is our honest read, as the people who put these models to work for Australian SMEs every week.
If you are still working out where AI fits in your business at all, start with our guide on how to use AI in your business, or the difference between an AI assistant and an AI agent. The model behind the agent matters less than picking the right workflow to automate first.
The smartest AI you can buy just got smarter, and twice as expensive. The winning move is not buying the biggest engine. It is putting the right engine on each job.
That is exactly how we build. TurnkeyAI agents run your routine workflows on the cost-effective Claude models, and reach for the heavy tier only when a job genuinely needs it. You pay frontier prices only where frontier intelligence earns its keep. You can compare the packages against your current monthly stack.
Claude Fable 5, answered straight.
Claude Fable 5 is the most capable AI model Anthropic has ever sold, released on June 9, 2026. It is the first public model from Anthropic's Mythos class, the frontier tier the company previously kept in the lab. On some of Anthropic's tests it scores more than 10% higher than the previous best model, Claude Opus 4.8.
It costs $10 USD per million tokens of input and $50 USD per million tokens of output, exactly double the previous top model, Claude Opus 4.8. In practice, having it read a 100-page document and write a 10-page summary costs around one US dollar. The same job costs about 50 cents on Opus 4.8 and about 10 cents on Haiku 4.5.
Mythos is the name of Anthropic's internal frontier tier, the models so capable the company previously chose not to sell them. Claude Fable 5 is the first model from that tier made available to the public, with new safety guardrails that made the release possible.
Anthropic shipped Fable 5 with new real-time safeguards. If someone asks about genuinely dangerous topics, like building weapons or attacking computer systems, the model blocks its own answer and a safer previous model responds instead. For normal business work you will never notice the guardrail.
Probably not for everyday automation. Routine workflows like quoting, enquiry handling and follow-ups run brilliantly on cheaper models, and paying double for them is wasted money. Fable 5 earns its keep on the genuinely hard jobs: large software projects, deep analysis, and reviewing very long documents in one pass.
Yes. The model is called claude-fable-5 and is available now through Anthropic's API and on Amazon Bedrock, for enterprise customers and paid subscribers.