Anthropic debuts cheaper Fable and Mythos AI models for demanding technical work

Complex engineering projects are the target of Anthropic’s latest AI release, with lower prices and a demonstration involving more than 1,000 agents. Announced November 9, Claude Fable 5.2 and Claude Mythos 5.2 share the same underlying model, but differ in safeguards and access for cybersecurity and biomedical work.

Fable 5.2 is scheduled for release Monday, November 9, Bloomberg reported. Mythos 5.2 is rolling out to selected organizations through verification programs, rather than receiving the same general release.

An eight-day operating-system experiment

Anthropic’s most striking coding example began with an empty folder and a request from one engineer to build a prototype operating system. The company says Fable 5.2 coordinated over 1,000 subagents—AI workers assigned parts of the job—for eight days of writing, testing and reviewing code. The resulting system boots on a laptop and includes disk, keyboard and network drivers, file storage, windows, a login screen and an installer.

Another internal project used agents running an earlier version of Fable 5.2 to move a critical production service from Python to Rust. Anthropic says the service became more efficient and reliable, with startup about 26 times faster. That example concerns a production migration; the operating system was an experiment.

The company’s evaluations also extend beyond software. On BenchCAD 2.1, which measures how closely hardware designs match reference geometry and circuits, Fable 5.2 scored 83.5 out of 100 at maximum effort. Fable 5.1 scored 44.4 and Opus 5.5 scored 65.3. These are Anthropic’s reported test results, not an independent assessment of engineering performance.

Faster output, but not always the first choice

Anthropic says the new models generate output up to 30% faster than their 5.1 predecessors, before the optional fast mode. Its pricing separates input tokens, the units processed in a request, from output tokens generated in response. The listed rates are lower than the previous versions, according to the announcement.

Effort settings change the cost-performance comparison. In Anthropic’s early FrontierCode 2.0 testing, Fable 5.2 at High effort scored 45.3%, close to Opus 5.5’s best score of 45.5%, for roughly half the cost per task. That benchmark measures whether code changes would be accepted by maintainers of open-source repositories.

Even so, Anthropic recommends starting with Opus 5.5 for most coding and knowledge tasks, then moving to Fable or Mythos when the work demands it. Fable can also support other models in two roles:

• Advisor: Haiku, Sonnet or Opus consults Fable to help solve harder problems.
• Orchestrator: Fable delegates simpler tasks to smaller models.

One underlying model, different permissions

The access split is deliberate. Fable has strong cybersecurity and biology safeguards intended to prevent misuse. Anthropic says those safeguards block most cybersecurity-related tasks. Organizations pursuing beneficial cybersecurity or biomedical research can apply for Mythos through its Life Sciences and Cyber Verification Programs, giving them a route to the same underlying model with fewer safeguards for those fields.

Anthropic says Fable 5.2’s cyber and biology protections are as robust as Fable 5.1’s while producing fewer unnecessary blocks. Its prerelease safety work included new evaluations for concerning behavior such as sandbox escapes—breaking out of a restricted computing environment. The company judges high-stakes risks from misalignment to remain low, describing a model acting contrary to intended goals.