Anthropic Brings Its Most Powerful Claude Model Yet to AWS with Built-in Safety Controls

Anthropic just dropped Claude Fable 5 on Amazon Bedrock and the Claude Platform on AWS. It’s the company’s first model to deliver what they call Mythos-class capabilities, paired with safeguards designed to let more businesses use it without the usual worries about misuse.

Fable 5 tops nearly every benchmark Anthropic tested. It’s built for heavy lifting—software engineering, research, and vision tasks that run for hours without needing a human to nudge it along. The model can chew through diagrams, charts, and tables buried in PDFs, making it a fit for finance, legal, analytics, architecture, and gaming teams. It also checks its own work against the original goals, catching mismatches before you do.

To keep things safe, Anthropic built in a fallback system. If a user sends a prompt that falls into high-risk areas like cybersecurity, biology, chemistry, or health, the model hands off to a less capable version called Opus 4.8. That lets Anthropic open up nearly all of Fable 5’s power to a broad audience. The full-strength version, Claude Mythos 5, is reserved for a small group of vetted customers.

On the technical side, getting started requires a few steps. You’ll need to opt into data sharing via the Data Retention API before you can invoke the model. Anthropic keeps inputs and outputs for 30 days for abuse detection. Pricing is tiered—harmful prompts that get routed to Opus 4.8 only cost you Opus rates. The model is live now in US East (N. Virginia) and Europe (Stockholm), with more regions on the way.

For developers, Anthropic provides SDK examples in Python and AWS CLI commands to hit the bedrock-runtime or bedrock-mantle endpoints. Console support isn’t here yet, but it’s coming soon.