Introducing Claude Fable 5.1 on AWS

At this time, we’re excited to announce the supply of Claude Fable 5.1 on Amazon Bedrock and Claude Platform on AWS. Claude Fable 5.1 delivers frontier intelligence for bold duties throughout coding, scientific analysis, and enterprise workflows.

Given its capabilities, Anthropic has designated Fable 5.1 a Coated Mannequin, a class of Claude fashions that carry further information retention, security evaluate, and entry insurance policies wherever they’re supplied. For extra data, see Amazon Bedrock abuse detection and Anthropic’s Coated Fashions web page. Anthropic additionally introduced the Enterprise Frontier Safeguards, an answer that may give prospects the choice to soundly deploy Anthropic’s most succesful fashions whereas conserving their information in cloud infrastructure they management. The Enterprise Frontier Safeguards part that follows covers what this implies on Amazon Bedrock.

This put up covers Claude Fable 5.1’s enhancements, the Enterprise Frontier Safeguards, and how one can begin constructing with the mannequin on Amazon Bedrock.

What makes Claude Fable 5.1 completely different

Anthropic reviews Fable 5.1 is a transparent enchancment over Fable 5 on the toughest reasoning assessments Anthropic runs. These assessments embody competitors arithmetic, graduate-level questions in engineering and the sciences, and lengthy multi-step issues the place reliability is vital. In day-to-day use that exhibits up as higher judgment on ambiguous duties and fewer assured incorrect solutions:

Agentic coding. Carries extra of a undertaking by itself, from code base-spanning options to code evaluate and efficiency work, throughout multi-hour classes. Additionally it is extra trustworthy: if it will get caught it says so, and it’s much less more likely to disable a failing take a look at to move.
Autonomous operation. Constructed for multi-hour jobs that span many purposes. It plans, makes use of the instruments it wants, recovers when a step fails, and retains you up to date with out being requested.
Finish-to-end data work. Takes an evaluation from first query to completed doc, doing the analysis, constructing the spreadsheet, writing the memo or deck, and checking its numbers because it goes. Constructed for on a regular basis finance, accounting, and healthcare work.
Scientific analysis. Helps analysis campaigns from literature and hypotheses to fashions, experiments, and formal verification.
Improved usability. Retains you up to date on lengthy duties, writes extra clearly, and follows directions extra intently.

Information retention

As a result of Claude Fable 5.1 is a Coated Mannequin, its use is topic to information retention for as much as 30 days and human evaluate by Amazon personnel. With Amazon Bedrock, you possibly can management information retention via a mode in your account or Amazon Bedrock undertaking, which you configure with the information retention API. Utilizing Claude Fable 5.1 requires the aws_review mode, which you set earlier than you invoke the mannequin. On this mode, AWS retains your prompts and outputs for human security evaluate inside the AWS boundary. Claude Fable 5.1 doesn’t require sharing with the mannequin supplier. See Amazon Bedrock information retention for extra particulars.

Enterprise Frontier Safeguards

Enterprise Frontier Safeguards (EFS), in-built partnership between AWS and Anthropic, will assist eligible prospects use Claude Fable 5 and Claude Fable 5.1 fashions whereas conserving their information in a cloud setting they management.

In case you are an EFS-eligible buyer, you need to use Claude Fable 5 and Claude Fable 5.1 with zero information retention (ZDR) on Amazon Bedrock and Claude Platform on AWS. That is out there for inside use via December 31, 2026. Further Enterprise Frontier Safeguards will probably be out there later this 12 months, it is possible for you to to maintain your prompts and outputs in your AWS account, underneath your personal encryption keys, entry insurance policies, and audit logging. Security monitoring will probably be via automated evaluate with no human evaluate required.

Getting began with Claude Fable 5.1 on Amazon Bedrock

To attempt Fable 5.1, open the Amazon Bedrock console, go to Check > Playground, and choose Fable 5.1 because the mannequin. From there, you possibly can run a immediate instantly towards it.

Programmatically, you possibly can name the mannequin with the Anthropic Messages API towards bedrock-runtime (via the Anthropic SDK). You can even keep on the Invoke and Converse APIs on bedrock-runtime via the AWS Command Line Interface (AWS CLI) and AWS SDK.

Conditions

Lively AWS account with Amazon Bedrock entry.
AWS Command Line Interface (AWS CLI) put in and configured.
Python 3.10+.
Boto3 put in: pip set up boto3.
AWS Id and Entry Administration (IAM) permissions: bedrock:InvokeModeland bedrock:InvokeModelWithResponseStream.

Right here’s a fast instance utilizing the AWS SDK for Python (Boto3):

import boto3
import json

# Create a Bedrock Runtime shopper
bedrock_runtime = boto3.shopper(
    service_name="bedrock-runtime",
    region_name="us-east-1"
)

# Invoke Claude Fable 5.1
response = bedrock_runtime.invoke_model(
    modelId="international.anthropic.claude-fable-5-1",
    contentType="utility/json",
    settle for="utility/json",
    physique=json.dumps({
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 4096,
        "messages": [
            {
                "role": "user",
                "content": "An S3 bucket serves 40 TB/month egress. Estimate the monthly egress cost at $0.09/GB, and state one architecture change to cut it. Show the calculation, keep it under 120 words."
            }
        ]
    })
)

consequence = json.hundreds(response["body"].learn())
# Fable 5.1 is a reasoning mannequin: the response could embody a pondering block
# earlier than the textual content block, so choose the textual content block reasonably than a set index.
print(subsequent(b["text"] for b in consequence["content"] if b["type"] == "textual content"))

You’ll be able to discover the Getting Began pocket book for extra examples.

Availability

Claude Fable 5.1 is offered as we speak on Amazon Bedrock via the US Geo CRIS (us.) and International CRIS (international.) inference profiles. In AWS GovCloud (US), it’s out there on each the bedrock-runtime and bedrock-mantle endpoints. See the Bedrock documentation for the total record of supported Areas. For pricing data, see Amazon Bedrock pricing. Additionally it is out there via Claude Platform on AWS in North America.

Give Claude Fable 5.1 a attempt within the Amazon Bedrock console, in Claude Platform on AWS, or discover the Getting Began notebooks on GitHub.