Summary: Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026, cutting the flagship price to $8 per million input tokens and $40 per million output tokens and adding an optional fast mode at up to 2.5x token generation speed. The two names are one underlying model, split into an all-users version and a verification-gated version. For builders running agentic coding or research workloads, the more interesting change is the effort system: at High effort, Fable 5.2 reportedly matches Opus 5.5’s best FrontierCode 2.0 score for roughly half the cost per task.
One model, two doors, again
Fable and Mythos remain the same underlying model. Fable ships with cybersecurity and biology safeguards, which is what lets Anthropic offer it to all users. Mythos drops those safeguards and is available only to organizations that apply through the Life Sciences and Cyber Verification Programs. I covered how this split worked in the 5.1 generation in One Model, Two Doors: Anthropic Splits a Frontier Release into Verified and Safeguarded Access, and the 5.2 release keeps that structure in place.
The safeguards themselves carry over with one adjustment worth noting: according to Anthropic, Fable 5.2’s cyber and biology safeguards are as robust as Fable 5.1’s but produce fewer unnecessary blocks. If your product sits anywhere near security tooling or biomedical data, the practical question is still whether your use case qualifies for Mythos or whether safeguarded Fable is enough. Anthropic also says it implemented its strongest protections yet against distillation attacks, and that it developed novel prerelease evaluations for behaviors like sandbox escapes.
The pricing and the speed numbers
The base pricing is $8 per million input tokens and $40 per million output tokens, down from the previous generation, and both models generate outputs up to 30% faster than Fable 5.1 and Mythos 5.1. The optional fast mode doubles down on latency: up to 2.5x faster token generation, priced at $16 per million input and $80 per million output.
That fast mode is a straightforward trade you can price per endpoint. If a user is waiting on a long agentic run, doubling the per-token cost to cut wall-clock time by up to 60% may be worth it for interactive flows and a waste for batch jobs. Anthropic does not publish guidance on when fast mode pays off, so you would need to measure your own latency-versus-token mix before committing.
Effort levels are the real cost lever
The pricing table is only half the story. Fable 5.2 runs at configurable effort levels, and the cost-per-task numbers change dramatically across them. On FrontierCode 2.0, which measures whether a model’s code changes would be merged by open-source maintainers, Anthropic reports that Fable 5.2 at High effort scores 45.3% versus Opus 5.5’s 45.5% at its best, for roughly half the cost per task. At Xhigh effort, it hits 49.1%, which Anthropic describes as the highest score of any released model it has tested.
The pattern repeats elsewhere. On Perplexity’s WANDR agentic search benchmark at Medium effort, Fable 5.2 matches Opus 5.5’s top score for about 40% of the cost per task. On Humanity’s Last Exam at Low effort, it matches Opus 5.5’s best for less than one-tenth of the cost. On Chartography chart reading at Medium effort, it scores 66.4% against Opus 5.5’s 64.4% at Max effort, again at less than a tenth of the cost.
The takeaway for anyone doing AI cost tracking across a model fleet: the effort setting you pick matters as much as the model name. Defaulting everything to maximum effort is how you pay premium prices for capability you did not need.
What the release claims about real work
Anthropic reports that over 90% of changes merged into its core codebase are now written by Claude with human guidance, and that a team of agents on an earlier Fable 5.2 version migrated a critical production service from Python to Rust, with the service starting about 26 times faster afterward. In one internal experiment, an engineer asked Fable 5.2 to design a prototype operating system from an empty folder; the model coordinated over 1,000 subagents over eight days and produced an OS that boots on a laptop, with its own kernel, drivers, file storage, and an installer. During prerelease testing, the model also formalized the proof of Fermat’s Last Theorem in 30 hours.
These are Anthropic’s own accounts, and it is worth being explicit about what they do and do not establish. The RoboDojo Sim robotics score was run internally, the benchmark tables are Anthropic-reported, and early tester quotes like Vercel’s Joe Haddad describing a model that understood intent without corrections are individual anecdotes. Independent replication of the cost-per-task claims does not exist yet, and the real economics of sustained agentic workloads at production scale are not demonstrated by any launch post.
Where it fits in your stack
Anthropic’s own recommendation is telling: start most coding and knowledge tasks with Claude Opus 5.5, the daily driver, and move up to Fable or Mythos only when scope and complexity demand it, or down to Sonnet 5.5 or Haiku 5.5 for simpler work. Fable 5.2 also supports two new roles in that hierarchy — Advisor, where smaller models consult it on hard problems, and Orchestrator, where it delegates simpler tasks to smaller models.
That framing tells you how Anthropic expects this to be used: not as a replacement for your tier setup, but as a new top of it. The practical next step is a small routing experiment. Take a handful of your real tasks, run them on Opus 5.5 at your current settings and on Fable 5.2 at Medium and High effort, and compare merged-quality output against actual cost per task. If the half-cost claim holds on your workload, the upgrade case writes itself; if it does not, you have learned your routing rules are already right and you have the data to prove it.
Sources
View the original post on X