Anthropic Unveils Fable 5.2, But Recommends Cheaper Model

Anthropic released Claude Fable 5.2 and Mythos 5.2 on November 9, five months after the first Fable 5 and Mythos 5 releases. The two versions share the same underlying model but operate under different access conditions, according to reporting by Elets CIO.

Fable 5.2 is available to all users, with safeguards targeting cybersecurity and biological risks. Mythos 5.2 is restricted to vetted organisations working in cyber defence and life sciences.

Fable 5.2 costs $8 per million input tokens and $40 per million output tokens. An optional fast mode doubles those rates to $16 and $80, respectively, while generating tokens up to 2.5 times faster.

The model is available through the Claude Platform and on Amazon Web Services, Google Cloud and Microsoft Azure.

Anthropic sells performance at a lower cost per task

Anthropic’s main selling point is not a lower token price but the amount of work the model can complete for the money.

The company says Fable 5.2 roughly matches Opus 5.5’s best performance on the FrontierCode 2.0 benchmark at about half the cost per task.

On FrontierSWE v2, a separate coding benchmark, Anthropic reports a score of 69.9 per cent. That compares with 62.3 per cent for Opus 5.5 and 65.5 per cent for OpenAI’s GPT-6 Astra.

Anthropic also says its safety classifiers block about 55 per cent fewer legitimate requests than the previous Fable 5.1 version.

However, the company is not positioning its newest model as the default choice for everyone. It continues to recommend Opus 5.5 for most coding and knowledge work.

Fable 5.2 is intended for the hardest open-ended technical tasks, with the ability to delegate simpler work to smaller models. That makes it a specialist option rather than a universal replacement for Anthropic’s existing models.

Mythos 5.2 targets security teams

For cybersecurity teams, Mythos 5.2 is the more specialised release.

In Anthropic’s own Common Vulnerabilities and Exposures (CVE) test, the model identified 10.2 per cent more known vulnerabilities than Opus 5.5.

Claude Enterprise customers can access it through Claude Security. Elets CIO also links the launch to CERT-In, India’s national cybersecurity agency, recently receiving access to Mythos.

The restricted distribution reflects the security implications of models capable of identifying software vulnerabilities. However, Anthropic has not opened Mythos 5.2 to general users.

Enterprise safeguards move monitoring into customer infrastructure

The release also introduces Enterprise Frontier Safeguards, an option designed for organisations with strict data-governance requirements.

The service runs safety monitoring on data within a customer’s own infrastructure, enabling zero-data-retention arrangements.

It is currently in private beta, with Bridgewater, Salesforce and Stripe among the participating customers.

The approach addresses a concern among enterprise buyers: how an AI provider can monitor potential misuse without retaining customer data. Anthropic’s proposed solution is to run the monitoring within the customer’s environment rather than its own infrastructure.

The private beta will provide an early test of whether that model can satisfy enterprise security and data-governance requirements.

The case for a multi-model strategy

Elets CIO draws a broader lesson from the release: organisations may no longer need one AI model for every task.

Its argument is that technology leaders should consider a mid-tier model for routine workloads, a frontier model for complex technical problems, and clear policies governing data retention and security monitoring.

That is an interpretation of the launch rather than a recommendation directly attributed to Anthropic.

For technology leaders evaluating the models, the distinction between token pricing and cost per completed task will be important. A more expensive model could prove cheaper for workloads that require fewer attempts, less human intervention or faster completion.

However, those savings will depend on the task, usage patterns and infrastructure costs.

All performance, pricing and safety comparisons cited here are based on Anthropic’s claims, as reported by Elets CIO, and have not been independently verified. Organisations considering adoption will need to test the models against their own workloads before determining whether the higher token price delivers better value.