Claude Fable 5.2 and Mythos 5.2 launch, with Mythos limited to verified orgs

Anthropic introduced Claude Fable 5.2 and Claude Mythos 5.2 on November 9. Fable 5.2 is open to all API customers at $8 per million input tokens; Mythos 5.2 has the same capabilities but is restricted to verified organizations.

In its post on X, the company’s Claude account described the pair as “designed to tackle the most daunting technical challenges” and as capable collaborators on ambitious projects in many fields. The post also claims improvements in areas such as coding. The text available to us is cut off at that point, and Anthropic’s release notes and docs include no benchmark scores to back the claim.

Same model family, two very different access rules

According to Anthropic’s Platform release notes, Fable 5.2 (claude-fable-5-2) is built for demanding reasoning and long-horizon agentic work. Mythos 5.2 (claude-mythos-5-2) is described as having the same capabilities, but it is available only to organizations verified through Anthropic’s programs, such as the Cyber Verification Program. Anthropic tells interested organizations to apply to the program that fits their use case or contact their account team.

Both models share a 1M token context window, up to 128k output tokens and adaptive thinking that cannot be switched off. Effort is controlled with a parameter that runs from low to max, with high as the API default. Anthropic advises most users to start with its latest Opus model and move to Fable 5.2 only when Opus at higher effort still falls short in their own evaluations.

Fable 5.2 runs on the Claude API, Claude Platform on AWS and Claude Models on Google Cloud, in both the Anthropic-managed and Google-managed versions. Mythos 5.2 is listed on the Claude API and the two Google Cloud variants. The documentation does not say whether either model is coming to Claude’s consumer plans.

Fast mode, set with speed: "fast" and no beta header, raises output speed by up to 2.5x. It costs $16 per million input tokens and $80 per million output tokens for both models.

Migration breaks some existing code

Anthropic lists four changes that can cause errors when moving from Fable 5.1:

• The check that detects edits to earlier conversation turns now applies to every account, and a mismatched thinking block returns a 400 error.
• The older computer_20251124 computer use tool also returns a 400 error. Developers must switch to the computer_toolset_20260801 toolset.
• In the beta advisor tool, a Fable 5.2 or Mythos 5.2 executor accepts only those two models as advisor.
• Forced tool use (tool_choice set to any or to a named tool) is not supported.

Thinking blocks produced by the new models are readable only by Fable 5.2, Mythos 5.2 and some older models in one direction. Moving a conversation to another model, including Opus 5.5 or Fable 5.1, drops that reasoning. Both models also require 30-day data retention and are unavailable under zero data retention unless Anthropic expressly authorizes it, which may rule them out for some regulated customers.

Twenty percent cheaper, but still a premium tier

The cost and economics picture is mixed. Fable 5.1 and Mythos 5.1 cost $10 per million input tokens and $50 per million output tokens, so the new $8 and $40 prices are a 20% cut, as our earlier pricing report also found. Anthropic’s pricing page puts cache reads at $0.20 per million tokens and batch processing at $4 input and $20 output, which halves the standard rates for work that can wait.

Even after the cut, Fable 5.2 sits well above the rest of the lineup. Based on our earlier reporting, Opus 5.5 costs $4 and $20 and Sonnet 5.5 costs $2 and $10. Fable 5.2 therefore costs twice as much as Opus 5.5 per token, and fast mode doubles that again. Anthropic’s own advice to try Opus first reads as a recognition of this gap: Fable is meant for the cases where the cheaper model fails.

Token price is also not the whole bill. Anthropic’s prompting guide says Fable 5.2 thinks before almost every reply at high effort and above, and that a line asking it to skip deliberation did not stop that at medium and above in the company’s testing. It adds that in coding products which invite delegation, subagents can account for a large part of what a session costs. Developers should expect real spend to depend on effort level and harness design as much as on the list price.

Mythos is priced the same as Fable, so the verification requirement, not money, is what separates the two. Anthropic does not explain in these documents what the verified tier is allowed to do that the open one is not, and it does not say what the safety classifiers on Fable 5.2 block. It does say that declined requests return a refusal, and that Opus 4.8 and Opus 5 are the permitted fallback targets.

What the announcement does not establish

The claims of stronger coding and broader capability come only from Anthropic. No independent testing or published scores accompany the launch, and the company does not give a knowledge cutoff in the release notes. Concrete early examples of the models’ work are limited to the company’s own showcase, such as the Diode electronics prototypes we covered earlier today. Whether the gap over Opus 5.5 justifies double the price will have to be shown by customers’ own evaluations.