Anthropic released Claude Fable 5.2 on Monday, November 9. It is the company’s newest generally available model, priced at $8 per million input tokens and $40 per million output tokens. That is twice the token price of Claude Opus 5.5, which costs $4 and $20.

The company that built Fable still tells you not to start there. On the announcement page, Anthropic writes that Fable is best for “the most ambitious, open-ended technical work.” For most coding and knowledge tasks, it says, “we recommend that users start with Claude Opus 5.5, our strongest daily driver,” and move up or down depending on the work. The model docs say the same: if you are unsure, start with Opus 5.5 for most workloads. Use Fable for heavy reasoning or for agent work that runs a long time, or when your tests on Opus still fall short, even if you give Opus more room to think.

The benchmark tables in that announcement put Fable ahead of Opus on many of the tests Anthropic ran. Those scores are Anthropic’s, and independent checks will take time, but even on its own charts the company still points most work at the cheaper model.

A shop in LA County does not buy the newest compressor for every service call. It should not run every voicemail, intake form, and quote draft through the most expensive model either.

The price gap across Anthropic’s current lineup is easy to see. Opus 5.5 is $4 and $20. Claude Sonnet 5.5 is $2 and $10, which Anthropic describes as the best combination of speed and intelligence. Claude Haiku 5.5 starts at $0.10 and $0.50 for prompts up to 100,000 tokens, and the docs list it for high-volume work that has to stay fast, such as classifying a request, pulling out details, and routing it. At those rates, the same short intake prompt costs about one-eightieth as much on Haiku as it does on Fable.

Most of the automation a trade or service business needs belongs on that cheaper end. Call summaries, lead intake, sorting a web form into the right pipeline, and drafting a standard quote from a price list need to be accurate and fast. They rarely need a model built for research agents that run for days.

An illustration, not a customer story

Picture a Glendale plumbing shop with two dispatchers and four trucks. The owner wants an assistant that transcribes after-hours voicemails, tags each one as emergency, routine, or a parts follow-up, and drafts a confirmation text. Once a week, a job comes in that is not on the price sheet: a commercial build-out with unusual fixtures and a tight inspection window. That one needs a careful quote and a second look from the owner.

If the vendor runs every voicemail through Fable, the shop pays the top-tier rate for work a fast, cheap model is built to do. A better setup uses a small, inexpensive model for the transcripts, tags, and routine texts, a mid-price model for ordinary quote drafts, and a stronger model only for the odd jobs that need judgment. The owner still reviews the unusual bids.

Ask what runs, and what it costs

When a vendor says the system uses the latest model, ask two questions. Which model runs which step? What does that step cost per task, on a typical week? Get the answers in writing. If intake, summaries, and routing all go through the same premium model, you are paying for capacity you do not use. If the vendor cannot tell you the model or the cost per task, you cannot tell whether next month’s bill is a busy week or a silent upgrade. How many times that model gets called matters too; we wrote earlier about running costs that jump when a job loops.

Kush AI Automation builds automations and AI agents for trades and service businesses around Los Angeles. We pick a model for each step, keep the expensive one for the work that needs it, and show you what a typical job costs to run. If you want that kind of setup on the tools you already use, talk to us or book a free call.

Sources