Anthropic releases Mythos 5.2 and Fable 5.2, saying they are "best for the most ambitious, open-ended technical work", for $8/1M input and $40/1M output tokens (Anthropic)

Anthropic has unveiled updated versions of Claude Mythos and Claude Fable, named Mythos 5.2 and Fable 5.2 respectively. These models are engineered to handle complex technical challenges, such as coding, reasoning, and vision tasks. During pre-release testing, they have assisted engineers in completing difficult software migrations, creating new electronic devices, solving problems in black hole physics, and making progress on robotics evaluations.

The Fable variant incorporates stringent cybersecurity and biology safeguards to prevent misuse, making it available to all users. Meanwhile, organizations interested in beneficial applications for cybersecurity or biomedical research can apply for access to Mythos through Anthropic's Life Sciences and Cyber Verification Programs.

The upgraded models communicate more clearly, generate outputs approximately 30% faster than their predecessors, and are priced more affordably at $8 per million input tokens and $40 per million output tokens. Additionally, an optional fast mode is available, providing up to 2.5 times faster token generation for $16 per million input tokens and $80 per million output tokens.

In terms of safety and alignment, extensive pre-release safety testing was conducted on the underlying model for Fable 5.2 and Mythos 5.2. This testing included developing novel evaluations to assess specific concerning behaviors, such as sandbox escapes. Current assessments suggest that high-stakes risks from misalignment are low, and the model's performance on most alignment measures is on par with or better than that of recent models.

Fable 5.2 and Mythos 5.2 are best suited for ambitious, open-ended technical work. For simpler coding and knowledge tasks, Anthropic recommends starting with Claude Opus 5.5, their most advanced daily driver. Depending on the project's scope and complexity, users can then transition to Fable or Mythos, or move down to Claude Sonnet 5.5 or Claude Haiku 5.5.

Notably, Fable 5.2 has demonstrated impressive engineering capabilities. An Anthropic engineer tasked the model with designing a prototype operating system from scratch, coordinating over 1,000 subagents that worked together for eight days to write, test, and review the code. The result was a functional operating system that boots on a laptop, featuring its own core, disk, keyboard, network drivers, permanent file storage, a login screen, and built-in Claude functionality.

Early testers have praised Fable 5.2's ability to apply strong engineering judgment to a diverse range of tasks.