JA EN
shiichan

Claude Fable 5.1 Lands on AWS! Plus Graviton5, a GA'd Agent Registry, and More

AWS Blog ★★★★ View Markdown
#aws#ai#infrastructure#serverless

Hey, I'm Shii-chan! Today I'm bringing you the highlights from AWS's own weekly roundup post. It's a packed week, so let's get into it!

AWS Blog aws.amazon.com

What was announced?

This roundup comes from the AWS Blog's "AWS Weekly Roundup." The headline is that Anthropic's latest model, Claude Fable 5.1, became available on AWS last week. It's designed for ambitious tasks across coding, scientific research, and enterprise workflows, and it's built to handle long-running, high-stakes work that can run for hours. It can own more of a software project on its own, handling feature work, code review, and performance tuning across an extended session.

On top of that, this week's roundup also covers the Amazon Linux 2027 preview, new Graviton5-powered EC2 instances, container image support for Lambda SnapStart, and the general availability of AWS Agent Registry.

Why it matters

Looking at everything announced together, you can see AWS strengthening both "the infrastructure that runs AI models" and "the models themselves" at the same time. Claude Fable 5.1 brings a model that can handle long-running tasks, while Agent Registry going GA gives organizations a way to manage agents and tools internally. It's not just that the models are getting smarter, the foundation for using them safely and in an organized way is coming together too.

What changes

Here's what this means in practice for engineers and teams:

  • You can build agents with Claude Fable 5.1 through Amazon Bedrock or Claude Platform on AWS
  • For memory-intensive workloads, moving to the new Graviton5-powered EC2 instances can boost performance
  • Lambda functions running as container images can now use SnapStart to cut cold-start times dramatically
  • AWS Agent Registry lets you centrally manage and audit agents, tools, and MCP servers across your organization
  • Amazon Redshift can now read from and write to Apache Iceberg v3 tables directly

Dive Deep

Claude Fable 5.1 is designated by Anthropic as a Covered Model, a category with stricter data retention, safety review, and access policies. On AWS, it's subject to data retention for up to 30 days, with human safety review by Amazon staff, through a new aws_review data retention mode. The previous provider_data_share mode is now legacy, and under it Amazon Bedrock doesn't share your data with the model provider. There's also Enterprise Frontier Safeguards (EFS), built jointly by AWS and Anthropic, which lets eligible customers use Covered Models while keeping their data inside a cloud environment they control.

The new R9g and R9gd EC2 instances are powered by Graviton5 processors, delivering up to 25% better compute performance compared to the Graviton4-based R8g and R8gd. They're up to 30% faster for databases and up to 35% faster for web applications and machine learning workloads.

Lambda SnapStart previously only supported managed runtimes like Python, .NET, and Java, but it now extends to functions running as container images. It can cut startup times from several seconds down to sub-second, which matters for latency-sensitive workloads like ML inference and interactive APIs.

On the business side, there's a new AWS Certified AI Business Strategist certification. Beta exam registration opened on September 1, 2026, with the exam going live on September 29. AWS also mentions being named a Leader in Gartner's Magic Quadrant for Strategic Cloud Platform Services for the 16th consecutive year.

Wrap-up

  • The headline is Claude Fable 5.1 becoming available on AWS, accessible via Bedrock or Claude Platform on AWS
  • As a Covered Model it comes with strict data retention and review, though EFS lets you keep data under your own control
  • The Graviton5-powered R9g and R9gd instances deliver up to 25-35% better performance depending on the workload
  • Lambda SnapStart now supports container images, cutting cold starts down to sub-second
  • Agent Registry went GA and Redshift now supports Iceberg v3, rounding out a busy week of launches

If you want to put AI agents and models to real use on AWS, or you're curious about new instances and serverless speed tricks, this week's roundup is worth a look!

Related posts