Anthropic's Claude Fable 5.2 arrives on Amazon's cloud
Anthropic, the company behind the Claude family of AI assistants, has released a new model called Claude Fable 5.2. Amazon announced on its machine learning blog that the model is now available through its cloud services.
What is Claude Fable 5.2?
An AI model is the brain behind a chatbot or assistant: the part that reads your request and writes a reply. Anthropic describes Fable 5.2 as its most capable generally available model for technical work, meaning ordinary customers can use it. The company says it raises the bar on the hardest problems in coding, maths and science.
It arrives alongside a sister model, Claude Mythos 5.2, which is not open to everyone. Mythos 5.2 is only available through Anthropic's verification programmes, which are for approved people doing advanced work in cyber security and life sciences research.
You can reach both models through two Amazon services. One is Amazon Bedrock, Amazon's online service for using AI models from different companies. The other is Claude Platform on AWS (Amazon Web Services, Amazon's cloud business).
What can it do?
Fable 5.2 is built to handle long, messy jobs without someone watching every step. According to the announcement, it can take on unclear, multi-step work, check its own output, catch its own mistakes and keep improving as it goes. It can run unattended for hours to days.
This kind of AI is often called an agent. An agent carries out a series of tasks by itself instead of answering one question at a time. Asking a colleague a quick question is like using a chatbot; handing them a whole project to finish while you are away is closer to using an agent.
Anthropic says Fable 5.2 suits these kinds of work:
- Writing and changing software across large collections of code
- Long-running agents that coordinate many smaller helper agents over several days
- Scientific research, from new ways of analysing data to formal mathematical proofs
- Financial analysis, including research and forecasting
- Design for 3D, hardware and motion, such as creating environments, parts and simulations in code
How reliable is it?
The announcement is careful about what it claims. In Anthropic's own testing, Mythos 5.2, the restricted sister model, made things up less often than several earlier Claude models. (When an AI confidently invents false information, people call it hallucination.) Mythos 5.2 also finished tasks more reliably, flagged important information more consistently and reported its own problems more honestly.
Those test results are for Mythos 5.2. For Fable 5.2, Anthropic says only that it expects the model to behave similarly, so the announcement gives no measured result for Fable 5.2 itself.
The model also has limits on what it will help with. It can sometimes decline requests about cyber security or biology. The announcement says developers can then try the request again on certain older Claude models. On Amazon Bedrock, though, that second attempt does not happen automatically, and a developer has to build it into their own software.
What happens to your data?
Small businesses should read this section carefully. Anthropic calls Fable 5.2 a Covered Model. By default, that means information you send it can be kept for up to 30 days and may be checked by people as part of a safety review. On Amazon Bedrock, Amazon staff carry out that review inside Amazon's own systems. Customers have to switch on a specific data setting for their account or project before they can use the model.
Anthropic and Amazon are building something called Enterprise Frontier Safeguards. It is meant to combine zero data retention, meaning your information is not kept, with tools for spotting misuse, by storing data in cloud systems that the customer controls.
It is not ready yet. Until it is, eligible customers can use Fable 5.2 with zero data retention for internal use until at least 30 June 2027. The announcement does not explain who counts as eligible.
How can people try it?
The simplest route is the Amazon Bedrock console, the website where AWS customers manage these services. Its Playground area lets you choose Fable 5.2 and type a request to it directly. You need an AWS account with access to Bedrock.
Some businesses already use older Claude models in their software. For them, Anthropic offers a helper for its coding tool, Claude Code, called the claude-api skill. It looks through existing software and guides developers through the changes needed to switch to Fable 5.2, such as removing old settings that no longer apply.
The skill is open source, which means its code is published for anyone to use. It is available on GitHub and comes built into several other coding tools.
The announcement does not list prices and points readers to Amazon Bedrock's pricing page instead. It also does not name every region where the model can be used, and directs people to Amazon's documentation for that.
If your business relies on technical work, or you pay developers who do, Fable 5.2 is worth knowing about. Make sure you understand its data rules before anyone sends it sensitive information.
Source: Claude Fable 5.2, Anthropic (announced via AWS Machine Learning blog). Listed in AI News Week 47, where we track every AI release each week.