JA EN
shiichan

Mythos-class power with safety built in: Claude Fable 5 lands on AWS!

AWS Blog ★★★★ View Markdown
#aws#ai#ai-safety

Hey there, it's me, Shiichan! Big news landed today. Anthropic's newest model, Claude Fable 5, is now available on AWS!

AWS Blog aws.amazon.com

What was announced?

Over on the AWS Blog, they announced that Claude Fable 5 is now available on Amazon Bedrock and Claude Platform on AWS.

The headline is that this model brings "Mythos-class" top-tier capabilities while shipping with strong safeguards from the start, so a much broader audience can use it safely. It's state-of-the-art on nearly all tested benchmarks, too. How cool is that!

Why it matters

Until now, the most cutting-edge models could also be misused, so only a limited group of people got to touch them. Fable 5 changes that by building in the brakes for risky requests from the start, opening up that top-class power to a much wider group of us.

What changes

Fable 5 shines in three big areas:

  • Extended task execution: It can keep running complex coding and knowledge-work tasks for long stretches without a human stepping in, the kind of work where earlier models would run out of steam.
  • Advanced vision: It reads diagrams, charts, and tables inside files and PDFs, so it can help across finance, legal, analytics, architecture, and gaming.
  • Self-improvement: It updates its own skills based on what it learns, and even builds its own harnesses and evaluations.

Dive Deep

Here's what the safeguards actually do. Harmful prompts around cybersecurity, biology, chemistry, and health don't get answered by Fable 5 itself; they fall back to a response from Opus 4.8 instead.

Harmful prompts related to cybersecurity, biology, chemistry, and health fall back to receive a response from Opus 4.8 instead.

By the way, the unrestricted version is called Claude Mythos 5, and it stays available only to a small group of vetted customers.

Let me line up where you can use it:

  • Amazon Bedrock: US East (N. Virginia) and Europe (Stockholm)
  • Claude Platform on AWS: North America, South America, Europe, and Asia Pacific

There's one condition to using it: you need to opt into Anthropic's data retention. Specifically, that means 30-day retention of inputs and outputs, plus human review. Pricing follows the standard Bedrock models, and fallback requests are billed at Opus 4.8 rates.

Access is gradually expanding for all AWS accounts based on your Bedrock usage patterns.

Wrap-up

Let's recap today's highlights:

  • Claude Fable 5 is now available on Amazon Bedrock and Claude Platform on AWS
  • The star of the show is "Mythos-class" capability plus built-in safeguards
  • Its strengths are extended tasks, advanced vision, and self-improvement
  • Prompts in risky domains fall back to Opus 4.8
  • Using it requires opting into 30-day data retention and human review

This one's for engineers who want to put a top-class AI to work on real tasks, and for teams that want that power with safety built in!

Related posts