Claude Fable 5.2 and Claude Mythos 5.2 are the same underlying model. Fable carries cybersecurity and biology safeguards and goes out to everyone; Mythos drops some of those and is limited to organizations that pass Anthropic's cyber or life-sciences verification. If you're running Claude inside a business workflow, Fable 5.2 is the one you'll actually use.
Pricing is $8 per million input tokens and $40 per million output. That's a 20% cut from Fable 5.1's $10/$50. There's also a fast mode at $16/$80 that generates output up to 2.5 times quicker. Here's the part worth noticing: $8/$40 is still exactly double Opus 5.5's $4/$20, which is the same 2x gap Anthropic set when the first Fable 5 launched in June at twice the price of Opus 4.8. The top tier got cheaper. The distance between the tiers didn't move at all.
The number most people will skim past
Anthropic published gains across the usual categories. FrontierCode 2.0 at xhigh effort went from 32.4% to 49.1%. Humanity's Last Exam with tools climbed from 65.6% to 74.2%. Chartography, a vision benchmark, went from 44.8% to 73.4%. Those are the lines that get quoted.
Further down the same table sits AutomationBench, Anthropic's business-workflow test: 31.4% on Fable 5.1, 51.4% on 5.2. Twenty points in one release, on the benchmark that lines up most directly with what a lead-routing agent, a document pipeline, or a support-escalation flow does all day.
A model can get much better at writing code and still be mediocre at finishing a multi-step business task without someone watching it. That's the gap this release closes.
Two other numbers point the same way. Remote Labor Index, a rough stand-in for "can it complete a real piece of knowledge work end to end," went from 17.9% to 41.3%. OSWorld 2.1, which measures computer use, went from 83.5% to 89.3%. Output generation is also about 30% faster than 5.1. Taken together the story isn't a smarter model so much as a more dependable one on long tasks.
Does doubling the rate pencil out?
Run a contract-review agent at ten contracts a day, 30,000 input tokens and 1,500 output tokens each. On Opus 5.5: (10 × 30,000 ÷ 1,000,000 × $4) + (10 × 1,500 ÷ 1,000,000 × $20) = $1.20 + $0.30 = $1.50 a day, roughly $45 a month. The same workload on Fable 5.2: $2.40 + $0.60 = $3.00 a day, about $90 a month.
For that particular job, the extra $45 probably buys you nothing. Contract review is reading comprehension and judgment, which Opus 5.5 already handles. Where the math flips is the workflow that half-works today — the agent that qualifies a lead, books the call, and updates the CRM, but drops the last step often enough that someone checks its work every morning. If the AutomationBench gain shows up on that task, $45 a month is the wrong figure to stare at. The right one is what that daily cleanup costs in salaried time.
Check the effort dial before you reach for the expensive model
Anthropic's own guidance is to start with Opus 5.5 for most work, step up to Fable when a task needs it, and drop to Sonnet 5.5 or Haiku 5.5 when it doesn't. Fable 5.2 can also run as an Advisor or Orchestrator — reviewing or directing a cheaper model's work instead of doing every step itself.
One efficiency note from the release complicates the simple "pay more, get more" read. At high effort, Fable 5.2 matches Opus 5.5 at max effort on FrontierCode 2.0 for roughly half the cost per task. Read that in reverse and it's a warning about the effort dial: plenty of "we need the frontier model for this" turns out to be "we needed a higher effort setting," which is the cheaper fix. Medium is the default in Claude Code and the Claude apps, so a task that's underperforming may never have been given room to think.
What's actually worth doing this week
Not a wholesale swap. The model ID is claude-fable-5-2 and it's live on the Claude API, AWS, Google Cloud, and Microsoft Azure, so changing it is a one-line edit — that was never the risky part. The original Fable 5 made that point in June, when a government directive pulled the model worldwide three days after launch with no migration window. Keep the model ID in a config variable and keep a fallback you've actually tested. Not because this release looks shaky, but because the last one proved how fast the ground can move.
The test that tells you something: take the two or three automations that still need a human to finish the last step, rerun them on Fable 5.2, and see whether the benchmark gain shows up on your work. If it does, the extra $40 or $50 a month almost always costs less than the hand-finishing it replaces. If it doesn't, you've spent an afternoon and learned your current tier is fine.
Would rather have someone run that comparison against your real workflow than extrapolate from a benchmark table? That's what an AI automation scoping call is for.
— Cole
Sources
- Anthropic — Introducing Claude Fable 5.2 and Claude Mythos 5.2 (November 9, 2026)
- Anthropic — Claude Platform pricing (model pricing table, fast mode pricing)