Claude Fable 5.2 jumps to 51.4% on AutomationBench and drops to $8 per million input tokens. Use it as the advisor, not the worker

Anthropic shipped a stronger, cheaper top model, and its own launch post tells you not to make it your default.

What happened

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9. They are the same underlying model: Fable carries cyber and biology safeguards and is open to all users, while Mythos goes to vetted organizations. Anthropic says the new models generate output up to 30% faster than the 5.1 versions and cost $8 per million input tokens and $40 per million output tokens. TechCrunch reports the input price fell from $10. An optional fast mode runs up to 2.5 times faster at $16 and $80.

On AutomationBench, the business workflow test in Anthropic's table, Fable 5.2 scored 51.4%, up from 31.4% for Fable 5.1. Opus 5.5 scored 42.5%, GPT-6 Astra 41.4%, and Gemini 4 Carbon 53.9%. Anthropic also says Fable 5.2 at High effort matched Opus 5.5's best FrontierCode score for roughly half the cost per task.

The most useful line is the guidance: for most coding and knowledge tasks, Anthropic recommends starting with Opus 5.5 and moving up or down as needed. Fable can act as an Advisor that smaller models consult on hard problems, or as an Orchestrator that hands simpler tasks to smaller models. It is available on AWS, Google Cloud and Azure.

My take

A 20 point jump on a business workflow benchmark is real progress, but read it plainly: the best models still pass about half of those tasks, and on this test a Gemini model scored higher. No single vendor wins everything.

For client automations, the advisor pattern is the part to copy:

1. Keep the high volume steps (classify, extract, route) on a small, cheap model.
2. Escalate only the ambiguous cases to Fable, with the full context attached.
3. Log every escalation so you can see what it costs per case and whether the answer changed anything.

At $40 per million output tokens, you want Fable making the judgment call, not writing every status email. And before switching any live workflow, rerun your own test cases. A benchmark tells you where to look. Your client's data tells you whether it works.