awsai mlPractitioner

Unlocking Mythos-Class Capabilities with Anthropic Claude Fable 5 on AWS

5 min read AWS BlogJun 9, 2026Reviewed for accuracy

Practitioner — Hands-on experience recommended

Anthropic Claude Fable 5 exists to elevate the capabilities of AI models, allowing you to tackle complex coding and knowledge work tasks that were previously unsustainable. This model represents a significant step forward, making Mythos-level capabilities accessible while ensuring safety through strong safeguards.

The core functionality of Claude Fable 5 lies in its ability to self-update skills based on learnings, effectively executing tasks for extended periods without needing human intervention. To access this powerful model, you must opt into data sharing using the Data Retention API by setting the provider_data_sharing parameter. This is crucial; without it, you won't be able to invoke the model. The model's self-sustaining nature means it can handle intricate requests, such as designing a distributed architecture on AWS that supports 100k requests per second across multiple geographic regions.

In production, be aware of a few critical details. First, if your AWS account doesn’t have access to Claude Fable 5 yet, it will be enabled soon based on your Bedrock usage. Also, remember that for Mythos-class models like Fable 5, Anthropic mandates a 30-day retention for all traffic. This can impact your data management strategies. Lastly, if a harmful prompt is mistakenly routed to Opus 4.8 instead of Fable 5, you'll only incur Opus pricing, which can be a cost-saving measure but also a potential risk if not monitored closely.

Key takeaways

→Opt into data sharing using the Data Retention API to access Claude Fable 5.
→Leverage the model's ability to handle complex tasks without constant intervention.
→Plan for a 30-day retention requirement for all traffic on Mythos-class models.
→Monitor your account access to ensure timely availability of Claude Fable 5.
→Be aware of potential cost implications when harmful prompts are misrouted.

Why it matters

This model can significantly enhance your team's productivity by automating complex tasks, allowing engineers to focus on higher-level design and architecture instead of routine coding.

Code examples

Bash

curl -X PUT https://bedrock-mantle.us-east-1.api.aws/v1/data_retention \
  -H "x-api-key: <your-bedrock-api-key>" \
  -H "Content-Type: application/json" \
  -d '{ "mode": "provider_data_share" }'

Python

1import anthropic
2
3client = anthropic.Anthropic(
4    base_url="https://bedrock-mantle.us-east-1.api.aws/anthropic",
5    api_key= <your-bedrock-api-key>
6)
7
8message = client.messages.create( 
9     model="anthropic.claude-fable-5", 
10	 max_tokens=4096, 
11	 messages=[ 
12	     { "role": "user", 
13		   "content": "Design a distributed architecture on AWS in Python that should support 100k requests per second across multiple geographic regions", 
14		 }, 
15	 ], 
16)
17
18print(message.content[0].text)

Python

1import boto3 
2bedrock_runtime = boto3.client("bedrock-runtime", region_name="us-east-1") 
3response = bedrock_runtime.converse( 
4    modelId="global.anthropic.claude-fable-5", 
5    messages=[ 
6        { 
7            "role": "user", 
8            "content": [ 
9                { 
10                    "text": "Design a distributed architecture on AWS in Python that should support 100k requests per second across multiple geographic regions." 
11                } 
12            ] 
13        } 
14    ], 
15    inferenceConfig={ 
16        "maxTokens": 4096 
17    } 
18) 
19print(response["output"]["message"]["content"][0]["text"])