Skip to main content

Fable 5.2 Got Cheaper, but Your Contract Picks the Model

Claude Fable 5.2 costs USD 8 and 40 per million tokens but requires 30-day retention. On a zero data retention contract, it may not be yours to use.

Fable 5.2 Got Cheaper, but Your Contract Picks the Model

Ricardo Argüello

On Monday, November 9, Anthropic released Claude Fable 5.2 and cut the price 20% against Fable 5.1. That is the part everyone repeated. The part that changes what you can use sits a few lines lower in the same documentation: Fable 5.2 requires 30-day data retention.

My thesis this week is short. If your company signed zero data retention with a vendor, your contract picks the model. The benchmark no longer gets a vote.

What Anthropic’s own pages say

I checked everything against the launch post, the API docs and the product pages, not against LinkedIn posts. The facts are these.

  • Fable 5.2 shipped on November 9, 2026, model ID claude-fable-5-2, on the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry.
  • It costs USD 8 per million input tokens and USD 40 per million output tokens. Fable 5.1 was 10 and 50.
  • Claude Opus 5.5 is 4 and 20, and Sonnet 5.5 is 2 and 10. Fable 5.2 costs double what Opus 5.5 does.
  • The model page says it “requires 30-day data retention and isn’t available under zero data retention unless expressly authorized by Anthropic”.

This is a per-model policy, not fine print in a plan. Anthropic calls Fable and Mythos Covered Models, versions 5, 5.1 and 5.2. Its reasoning is that some misuse only shows up across many requests, so the safeguards need retained data. Prompts and responses are kept at least 30 days and then deleted automatically, unless a safety investigation or a legal duty applies. Other Claude models stay under your existing agreement and settings.

Zero retention: two texts that disagree a little

Read this part slowly.

The launch post says eligible customers can keep using Fable 5.2 with zero data retention before adding Enterprise Frontier Safeguards (EFS). EFS runs the safety monitoring on prompts and responses stored in the customer’s own infrastructure. Everyone else gets 30 days.

The support article narrows that exception. It describes temporary ZDR for internal business applications, as a bridge to EFS, which rolls out in phases starting fall 2026. It adds that Anthropic may modify or withdraw the arrangement.

Read it the way a lawyer would. A temporary exception the vendor can withdraw is not a right. If a critical process leans on it, your contract leans on someone else’s decision. I wrote about the switch you don’t control when a government shut off two models over a weekend. This is the same risk with a date on the calendar.

One more thing. On Amazon Bedrock and Google-managed Google Cloud, the docs say the cloud provider is the data processor and send you to its terms. The Covered Models article says the policy applies on every platform. I wouldn’t settle that gap myself. Ask your cloud provider in writing.

Claude Meetings, Sheets and Whiteboards

The week had more. On November 10 came Claude Sheets and Claude Whiteboards (the LinkedIn posts that circulated call it “Claude in Sheets”). On the 12th, Claude Meetings.

Meetings has concrete details. It is in beta on Team and Enterprise plans, and it saves a transcript, a summary and screen descriptions, not audio. By default it stays out of calls with outside guests, and an admin can allow them. And it is not available to organizations with zero data retention, a HIPAA-ready configuration or customer-managed encryption keys.

What those pages don’t say: how long the transcript is kept, how it gets deleted, and which model processes it. The Sheets and Whiteboards announcement doesn’t give retention or plan availability either. I won’t guess. If you plan to switch Meetings on, those three questions go in an email to your account rep before the pilot.

That catches my attention. Twice in one week, the organizations with the strictest protection ended up outside the new thing.

What I’d do before choosing a model

Here is the order I’d follow, in four steps.

  1. List the workflows that run on Anthropic models today and mark which ones touch data under a zero-retention clause, yours with a customer or your sector regulator’s.
  2. For those, the model question is already answered: Opus 5.5, Sonnet 5.5, or any model that isn’t a Covered Model. At half the price of Fable, too.
  3. For workflows that can accept 30 days, test Fable 5.2 against Opus 5.5 on your own tasks. Paying double only makes sense if your measurement shows it. The benchmarks Anthropic publishes are measured by Anthropic.
  4. Ask your vendor in writing what retention applies to your account and on which platform, and keep the answer.

What IQ Source does

With Tech Partner we put the model choice in a configuration layer, so moving a workflow from Fable to Opus is a one-line change, not a project. Before that, IQ Source Maestro sorts the workflows by the retention they can accept. Commercial terms are text you sign, and in this look at what Anthropic published versus what you sign I separated what is written from what is intent.

I’ve been programming since 1990 and I use Claude Code every day. I’ll take the model my contract lets me use over one that forces me to renegotiate with my customers. Check today what your retention clause says. If you can’t find it, that’s your first to-do.

Check which model your retention contract allows

Frequently Asked Questions

How much does Claude Fable 5.2 cost and how long does Anthropic keep my data?

Claude Fable 5.2 costs USD 8 per million input tokens and USD 40 per million output tokens. Anthropic designates it a Covered Model and requires 30-day retention of prompts and responses for safety monitoring. Under zero data retention it is not available without Anthropic's express authorization.

Can Claude Fable 5.2 be used with zero data retention?

Only with Anthropic's express authorization. The API documentation says Fable 5.2 requires 30-day retention. The November 9 announcement says eligible customers can keep zero retention while Enterprise Frontier Safeguards arrives, and the support article describes that as a temporary transition Anthropic may withdraw.

What does Claude Meetings store from a call, and does it work with zero data retention?

Claude Meetings saves a transcript, a summary and screen descriptions, not audio. It is in beta on Team and Enterprise plans and is not available to organizations with zero data retention, a HIPAA-ready configuration or customer-managed encryption keys. Anthropic does not state how long it keeps that data.

How do I choose between Claude Fable 5.2 and Claude Opus 5.5 for my company?

Start with the retention each workflow can accept, before looking at benchmarks. Claude Fable 5.2 costs USD 8 and 40 per million tokens and requires 30-day retention. Claude Opus 5.5 costs USD 4 and 20 and is not a Covered Model. If a workflow handles zero-retention data, the choice is already made.

  • Claude Fable 5.2
  • Anthropic
  • data retention
  • zero data retention
  • Claude Meetings
  • AI governance
  • model selection

Newsletter

Get the best articles in your inbox

We respect your privacy. Unsubscribe anytime.