Anthropic发布Claude Fable 5.2与Mythos 5.2模型

We're introducing Claude Fable 5.2, Claude Fable 5.2 Fast, Claude Mythos 5.2, Claude Mythos 5.2 in Microsoft Foundry, expanding what teams can tackle with AI across software engineering, scientific research, and finance.

These are fields where the hardest work starts with an open question. An engineering team needs to find a root cause across several repositories. A researcher needs to test an approach and determine whether the results hold up. A financial analyst needs to uncover an edge case that changes the conclusion. Fable 5.2 and Mythos 5.2 bring new capabilities to work that depends on judgment, sustained investigation, and careful verification.

Fable 5.2 takes on bigger technical challenges

Claude Fable 5.2 and Fable 5.2 Fast are Anthropic's most capable generally available models for technical fields. It can take on large, loosely defined software projects, work across repositories and mono repos, and apply architectural judgment while testing its changes. Its willingness to revisit assumptions helps it find root causes that previous Claude models missed.

That ability extends to leading longer agent workflows. In Anthropic's testing, Fable 5.2 worked through ambiguous engineering problems for hours and up to days, coordinating subagents and flagging results that could not be reproduced. For developers, this opens up more ambitious tasks to delegate, with the model carrying work through successive rounds of implementation and verification.

The same approach supports scientific and financial work. Fable 5.2 can help formalize mathematical proofs, develop methods for difficult physics computations, and investigate subtle edge cases in quantitative financial analysis.

Mythos 5.2 brings specialized capabilities to cybersecurity and life sciences

Claude Mythos 5.2 and Mythos 5.2 Fast are Anthropic's most capable models for security engineering and life sciences. For authorized security teams, it supports challenging work such as hardening systems and understanding undocumented software. In Anthropic's testing, it improved on previous Claude models in building defenses and analyzing undocumented binaries for potential vulnerabilities.

For life sciences teams, Mythos 5.2 can help synthesize literature, orchestrate analyses, and support basic and translational research, with particular strength in biomolecular design. Access is limited to authorized customers through Anthropic's verification programs.

Pricing

Model

Deployment Type

Input/M Tokens

Output/M Tokens

Cache Read

Cache write

Batch I/P

Batch O/P

Availability

Global Standard,

US DataZone

GA, Hosted on Anthropic

Global Standard,

US DataZone

NA

NA

GA, Hosted on Anthropic

Stronger alignment and safeguards

Anthropic ran its alignment testing on Mythos 5.2, the same underlying model as Fable 5.2 without Fable 5.2's cyber and bio safeguards. Mythos 5.2 scored lower than Opus 5.5, Sonnet 5.5, and Mythos 5.1 on behaviors including hallucinating inputs, falsely claiming to have completed a task, and omitting important information. Anthropic expects Fable 5.2 to behave similarly. Anthropic has also included safeguards intended to reduce misuse in cybersecurity, life sciences, and distillation. Where a cybersecurity or biology query is blocked, customers instead receive a response from Opus 4.8 for cybersecurity and Opus 5 for biology. Fable 5.2 and Mythos 5.2 are covered models, which require data retention for safety monitoring. Until Enterprise Frontier Safeguards is available, eligible customers can use Fable 5.2 with zero data retention.

Explore Claude Fable 5.2 and Claude Fable 5.2 Fast in Microsoft Foundry for your most demanding technical projects. For Claude Mythos 5.2, contact your Microsoft account team to confirm eligibility and access.