A new model, a new question
Anthropic has launched Claude Fable 5 — described as its most capable, widely released model to date, built for the most demanding reasoning and agentic work. That's a genuine milestone. But the most interesting part of the launch isn't how capable the model is. It's that it costs twice as much as Anthropic's previous flagship — which forces a question most businesses don't ask often enough: do you actually need the newest, most capable model — or just one that's good enough?
We wrote about this when Claude Opus 4.8 launched: model choice isn't a technical detail for IT to quietly sort out. It's a governance decision, on par with who has access to what. Fable 5 is a good example of why.
What Claude Fable 5 actually is
Fable 5 is built for the hardest tasks: long, autonomous agentic workflows, complex chains of reasoning, and work where the model needs to spend minutes — not seconds — thinking before it delivers an answer. A few practical characteristics:
- A 1 million token context window by default — the model can hold a very large amount of information in context at once.
- Thinking is always on. Unlike other Claude models, Fable 5 can't turn off its reasoning step. It always deliberates carefully before answering, which improves quality on complex tasks but also means longer response times.
- Built for long, autonomous work — first-shot implementations of complex systems, deep code review, and tasks where the model coordinates multiple sub-agents on its own.
This isn't a model built to draft a faster email. It's a model built for tasks that today take a skilled senior employee several hours to do well.
The price tells the real story
Here's the number that actually matters for your decision: Fable 5 costs $10 per million input tokens and $50 per million output tokens. For comparison, Claude Opus 4.8 — Anthropic's previous flagship, still highly capable — costs $5/$25. Fable 5 is exactly twice the price.
That's not an accident. It's a signal that Fable 5 is a specialised model for specialised needs, not an upgrade everyone should default to. Using it for tasks Opus 4.8 or Sonnet 5 already handle well is money out the window — the same point we've made about Copilot licences nobody uses correctly.
When does Fable 5 actually make sense?
For the vast majority of Norwegian SMBs, the answer is: not yet, and perhaps never at scale. But it's worth knowing when it becomes relevant:
- Long, complex agentic assignments — for example, an agent that independently carries out a multi-step migration or a large piece of analysis over an extended period, without human intervention along the way.
- First-attempt implementations of well-specified systems — where the quality of the first draft matters more than response time.
- Deep code review and debugging in large, complicated codebases.
- Low-quality images or degraded documents — Fable 5 is trained to handle visually demanding input better than earlier models.
Notice the pattern: these are tasks where the quality of a single, complex answer is worth far more than the cost — not high-volume, everyday tasks. For most businesses, Sonnet 5 or Opus 4.8 still covers day-to-day needs well.
One governance detail worth noting
Here's something that genuinely belongs in a governance conversation, not just a technical comparison: Fable 5 requires a minimum of 30 days of data retention with Anthropic. It isn't available to businesses that have agreed to zero data retention.
For most Norwegian SMBs, this isn't a dealbreaker — but it's exactly the kind of detail an AI Ready assessment should catch before you adopt a new model, not after. If data retention is part of your privacy assessment (DPIA), this needs to be evaluated explicitly, not discovered by accident.
A brief note on Claude Mythos 5
Anthropic also has a sister model, Claude Mythos 5 — the same capability and pricing as Fable 5, but available exclusively through Project Glasswing, Anthropic's defensive cybersecurity coalition. Unless your business is part of that initiative, Fable 5 is the one that's relevant. We mention it mainly because the names are easy to confuse.
The point that keeps repeating
The newest model isn't automatically the right model. Fable 5 is a genuine step forward for the tasks it's built for — but switching to it everywhere "because it's newest" is exactly the kind of unstructured decision we keep warning against. The right model choice, like the right access control, should be made deliberately — not defaulted to the most expensive option just because it exists.
AI starts not with technology. It starts with structure.
Get in touch for a free AI Ready assessment →
Read also: Claude Opus 4.8 Is Here — But the Model Isn't Your Bottleneck · You're Using AI to Save Time. That's Step One of Five