Skip to content

AI & Automation1 min read

Anthropic releases specialised Claude models

Anthropic has launched Claude Fable 5.2 and Mythos 5.2 to target technical automation tasks.

by Fini MironFounder @ Balkaris
An abstract composition printed flat in pale cool grey, steel blue, deep navy and bright mustard yellow: one tall narrow column standing on the bottom edge and reaching almost to the top, with three short horizontal bars crossing it at uneven heights, the rest of the frame one flat empty field.
Drawn for this article on our own machine. It is not a photograph.

Anthropic has released Claude Fable 5.2 and Mythos 5.2, according to a report by The-decoder. These new models aim to reclaim the performance lead in artificial intelligence ahead of the company's planned IPO. The Fable version includes safety restrictions for cybersecurity and biology, while Mythos is restricted to verified organisations. Anthropic claims these models produce output up to 30 percent faster than their predecessors. The company continues to recommend Claude Opus 5.5 for general, everyday use.

Why it matters

The release of these models shifts the focus from general chat to specialised, high-cost technical automation. For companies integrating AI agents into existing workflows, the choice between models now involves a direct trade-off between safety compliance and raw technical capability. You must account for the fact that peak performance comes with significantly higher per-task costs and specific platform requirements.

  • Budgeting must reflect higher costs for specialised technical tasks compared to general models.
  • Security-sensitive workflows may require the Mythos variant through formal verification programmes.
  • Implementation speed may increase if you use supported platforms like Google Cloud or Microsoft Foundry.

What to watch

You should audit your token budgets before deploying these models. While Anthropic has lowered the price per million tokens, the actual cost per task is roughly 2.6 times higher than using Opus 5.5. If your business requires agentic coding or terminal-based workflows, compare these benchmarks against competitors, as Fable 5.2 currently trails in those specific areas.

Within a year, specialised models will become the standard for high-stakes tasks. This requires the cost per token to drop by at least 70% through hardware optimisation.

Key takeaways

  • Specialised models increase the cost per task compared to general models.
  • Technical workflows now require a choice between safety restrictions and raw performance.
  • Hardware optimisation must reduce token costs by 70% to make these models standard.

Frequently asked

How do I decide between Fable 5.2 and Mythos 5.2?

It depends on your organisation's verification status and whether you require specific safety restrictions for biology or cybersecurity.

Is it cheaper to use these new models for all tasks?

No. Using these models is roughly 2.6 times more expensive per task than using Claude Opus 5.5.