Claude Mythos 5.2 and Fable 5.2 Arrive Faster and at a Lower Price

Claude Mythos 5.2 arrived this Monday, November 9, as Anthropic's new frontier version. For the general public, the same model appears under the name Fable 5.2, with stricter safeguards.

The results released point to superiority over Opus 5.5 in almost all synthetic benchmarks. However, the areas cited include coding, research, cybersecurity, and general knowledge.

Price and speed changed at launch

Two practical changes stand out.

First, speed. In addition, testers report responses up to 30% faster than the 5.1 editions.

Second, cost. Input costs US$8 per million tokens, while output costs US$40 per million.

There is also an optional mode called Fast. However, it makes generation 2.5 times faster, but at double the price.

The demo that shows Claude's new role

The company used Fable 5.2 to develop an operating system prototype.

The model coordinated 1,000 smaller agents over eight days. They wrote, tested, and reviewed the code.

The result stands out for its scope. In addition, the outcome was a functional system capable of running on a laptop, with its own kernel and drivers.

The prototype also includes window management and integration with the assistant itself.

This capability is organized into two declared roles.

In the Consultant role, models such as Haiku, Sonnet, or Opus consult it to solve more complex problems.

As Orchestrator, it delegates simpler tasks to smaller models.

Where it wins and where it loses in the tests

The comparison involved Fable 5.1, Opus 5.5, GPT-6 Astra, and the recent Gemini 4 Carbon.

The new model outperforms the others in most of the evaluations presented.

However, two exceptions deserve mention. Still, it falls behind Opus 5.5 in the knowledge-work test, AA-Briefcase v1.1.

Gemini 4 Carbon, in turn, wins the enterprise workflow test, AutomationBench.

The company itself advises choosing based on complexity. According to it, Opus 5.5 remains the best option for everyday use.

The internal data on AI-written code

However, this number helps explain the bet.

More than 90% of the changes added to the company's codebase are written by the assistant, always under human supervision.

The new model entered this workflow with specific functions. It resolved more persistent bugs quickly, optimized service performance, and completed higher-value software projects.

On the security front, Mythos 5.2 appears as the most capable for cybersecurity engineering. The company states that it finds more vulnerabilities and performs better at reproducing and resolving issues.

How to access each version

Access remains split.

Mythos 5.2 remains restricted. However, it is available to participants in the Cyber Verification Program and the Life Sciences Verification Program.

Corporate clients can also consult it through Claude Security.

Fable 5.2 reached the general public across all of the assistant's channels. In addition, it is also available on the Amazon Web Services, Google Cloud, and Microsoft Azure platforms.

On the native platform, the identifier is claude-fable-5-2.

What to evaluate before switching in your project

First, compare the cost per completed task. A model that is more expensive per token can turn out cheaper when it solves tasks in fewer iterations.

Second, test the orchestrator role in your workflow. Delegating simple tasks to a smaller model tends to yield significant savings.

Third, measure the Fast mode only where latency matters. In addition, doubling the price makes sense in interactive scenarios, less so in batch processing.

Finally, note the benchmarks that diverge. Knowledge work and enterprise automation had different winners, so your use case decides.

Follow our profile on Instagram!