Claude Fable 5.2 on Amazon Bedrock: Price, Regions, Data Retention and How to Use It

Claude Fable 5.2 is the newest model in Anthropic's Fable line, positioned above the Opus models. AWS's model card describes it as "Anthropic's frontier model for technical work, setting a new bar in coding, finance, and the sciences" and says it "takes on ambiguous problems for days, questions and verifies its own work, and orchestrates subagents to optimize results in a loop" (AWS model card).

Amazon's newsroom post says it is built for long-running autonomous work and that it "runs for hours to days unattended, coordinating subagents and verifying its own output." It adds that in Anthropic's testing, Fable 5.2 acted as a lead agent coordinating hundreds of subagents on multi-day tasks (About Amazon).

The key specifications, as published by AWS and Anthropic:

Sources: AWS model card, Anthropic model overview.

What about Claude Mythos 5.2?

Mythos 5.2 is the same underlying model without Fable 5.2's cybersecurity and biology safeguards. Anthropic offers it only to organisations verified through programmes such as its Cyber Verification Program, and it shares Fable 5.2's specs and pricing (Anthropic). On AWS, Bedrock also made Mythos 5.2 available to allow-listed customers in AWS GovCloud (US) on 9 November. AWS calls it the first Anthropic model with strong cybersecurity capabilities in GovCloud (AWS What's New).

How much does Claude Fable 5.2 cost on Amazon Bedrock?

Anthropic's list price for Fable 5.2 is $8 per million input tokens and $40 per million output tokens. Fable 5.1 was $10 and $50, so that's a 20% cut (Anthropic). Anthropic's own price list also includes:

Source: Anthropic model overview and Claude Fable page.

Important caveat for AWS buyers: AWS's model card sends you to the Amazon Bedrock pricing page for Bedrock rates. When we checked on 10 November, that page didn't yet show a Fable 5.2 row we could read. We haven't confirmed the Bedrock price is identical to Anthropic's, so check the console or pricing page before you budget. Batch and fast mode in the table above are Anthropic API features. The Bedrock model card lists only the Standard service tier for Fable 5.2: Priority, Flex and Reserved are not supported.

What does a typical request cost?

Estimate, at Anthropic's list price: a request with 100,000 input tokens and 10,000 output tokens costs about $0.80 + $0.40 = $1.20. Keep in mind that Fable 5.2's thinking is always on, and thinking tokens count as output. At high or max effort, output can be much larger than the visible answer. One Hacker News user who ran a public SVG benchmark reported that Fable 5.2 at high effort "wrote 76,158 tokens for $3.05" while Opus 5.5 "wrote 19,767 for $0.40" on the same task (Hacker News). That's one user's test, not an official figure, but it shows why token prices alone can mislead.

Fable 5.2 vs Opus 5.5 vs Sonnet 5.5 vs Haiku 5.5: which should you use?

This is the decision most Bedrock teams actually face. Prices below are Anthropic list prices per million tokens.

Source: Anthropic model comparison. For more on the cheapest tier, see our Claude Haiku 5.5 on Amazon Bedrock guide.

A simple rule: start on Sonnet 5.5 or Opus 5.5. Move a workload to Fable 5.2 only when you can show the cheaper model failing at it, for example an agent that loses track halfway through a long task. Fable's thinking can't be switched off, so it's the wrong choice for chatbots or short, latency-sensitive calls.

Which AWS Regions support Claude Fable 5.2?

Fable 5.2 has no in-Region endpoint anywhere. You call it through a cross-Region inference profile:

• Geo (US) profile, us.anthropic.claude-fable-5-2: source Regions us-east-1, us-east-2, us-west-1, us-west-2, ca-central-1 and ca-west-1. Requests are routed within the US (plus the Canadian Region when you call from Canada).
• Global profile, global.anthropic.claude-fable-5-2: callable from 30+ commercial Regions, including Frankfurt, Ireland, London, Paris, Stockholm, Tokyo, Seoul, Mumbai, Singapore, Sydney, São Paulo, UAE and Cape Town. Requests can be routed to any Region worldwide.
• GovCloud (US-East and US-West): Geo profile only (AWS What's New).

Source: AWS model card regional table.

What this means outside the US: the model card lists no EU or APAC Geo profile for Fable 5.2. If your company needs inference to stay inside the EU, Japan or Australia, the only route today is the Global profile, and that routes worldwide. Our Bedrock data residency explainer covers how in-country and Geo profiles normally work.

Why does Fable 5.2 need data retention on Bedrock?

Anthropic has designated Fable 5.2 and Mythos 5.2 as "covered models" that require data retention for safety monitoring (AWS What's New). In practice:

1. You must opt in before the first call. The model card says: "To use this model, you must opt in to AWS review by setting your data retention mode to aws_review via the Data Retention API" (AWS).
2. Retention is up to 30 days, with human review. AWS's launch blog says use "requires retention for up to 30 days and human safety review," and that on Bedrock "Amazon personnel perform the review within the AWS boundary" (AWS Machine Learning Blog).
3. There's a zero-data-retention exception. Until a product AWS and Anthropic call Enterprise Frontier Safeguards (EFS) is available, "eligible customers can use Fable 5.2 with zero data retention." AWS's blog says that applies to internal use through at least 30 June 2027. AWS doesn't spell out eligibility in the announcement, so ask your account team.

If your legal team has signed off only on zero-retention AI use, raise this with them before anyone enables the model.

What happens when Fable 5.2 refuses a request?

Fable 5.2 has classifiers that block high-risk cybersecurity activity and most research-biology tasks. On Bedrock, a blocked request returns HTTP 200 with stop_reason: "refusal", not an error, and AWS tells customers to treat that as a "primary response path" (AWS model card).

AWS's blog says cybersecurity queries can be retried on Claude Opus 4.8 and biology queries on Claude Opus 5, but on Bedrock your code must do the retry. A beta "fallback credit" token, which some refusals include, lets the retry bill the cached prefix at cache-read rates instead of cache-write rates. Security vendors and life-sciences teams should budget for that fallback path, or apply for Mythos 5.2 access.

How to use Claude Fable 5.2 on Amazon Bedrock (step by step)

1. Check permissions. Your IAM role needs bedrock:InvokeModel and bedrock:InvokeModelWithResponseStream (AWS blog).
2. Set the data-retention mode to aws_review for your account or project, after checking with legal (see above).
3. Pick a profile. Use us.anthropic.claude-fable-5-2 if you need US-only routing, or global.anthropic.claude-fable-5-2 otherwise.
4. Try it in the console. In the Bedrock console, go to Test → Playground and select Claude Fable 5.2.
5. Call it from code. Bedrock supports the Anthropic Messages API on bedrock-runtime, as well as Converse and Invoke. AWS's own Converse sample (from the model card) looks like this:

import boto3

client = boto3.client('bedrock-runtime', region_name='us-east-1')
response = client.converse(
    modelId='global.anthropic.claude-fable-5-2',
    messages=[
        {
            'role': 'user',
            'content': [{'text': 'Can you explain the features of Amazon Bedrock?'}]
        }
    ]
)
print(response)

6. Watch out for the breaking changes. Per AWS and Anthropic: temperature must be 1.0 or unset, top_p 0.99 or unset, and top_k isn't supported. Forced tool use (tool_choice of any or tool) returns a 400 error. Thinking blocks must be replayed unchanged in multi-turn conversations, or the request fails with a 400 (Anthropic).
7. Know what's missing. The Bedrock model card lists Count Tokens, Intelligent Prompt Routing and implicit prompt caching as not supported. Explicit prompt caching works, with a 512-token minimum per checkpoint, up to four checkpoints, and 5-minute or 1-hour TTLs.

What are developers saying about Fable 5.2?

The main Hacker News launch thread passed 800 comments within a day, and opinion is split. All quotes below are from that thread and reflect individual users, not tests by 9to5Amazon:

• On price, one commenter welcomed the cut: "The biggest relief is that the price has dropped by 20%. It's still expensive…" (kimseungyong).
• Others weren't convinced it's worth double Opus 5.5: "it makes the same exact errors as Opus, but slower and much more expensive" (user43928).
• A warning for agent builders: one user asked for "max 4 subagents" and said the model "decided to spawn 6+ nested subagents," draining their plan's five-hour usage allowance in under 20 minutes (TuxSH). That was on a Claude subscription, not Bedrock, but on Bedrock the same behaviour would show up as token spend.
• A cost-per-task view: one commenter argued effort level matters more than list price, saying medium and high effort "cost $3.80 and $9.08 respectively" on a public benchmark (kingstnap).

We didn't find Bedrock-specific Reddit threads yet. The questions readers are searching include "Claude Fable 5.2 vs Opus 5.5", "Claude Fable 5.2 price" and "Claude Fable 5.2 benchmarks".

Editor's take

Fable 5.2 matters to AWS customers less for its benchmarks than for the access terms around it. A 20% price cut, a 1M-token context and same-day Bedrock availability are good news. But the mandatory aws_review retention mode, the lack of any in-Region or EU/APAC Geo endpoint, and refusal handling you have to build yourself make it a model to adopt deliberately, not by default. For most teams, the sensible move is to keep Opus 5.5 or Sonnet 5.5 as the workhorse and send only the hardest long-horizon jobs to Fable. That fits the wider Amazon–Anthropic relationship we covered in our deal explainer. If you're also rethinking which AWS AI tools your business teams use, see today's piece on Amazon Quick becoming AWS Quick.

FAQ

Is Claude Fable 5.2 available on Amazon Bedrock?

Yes. It became available on Amazon Bedrock and Claude Platform on AWS on 9 November 2026, and in AWS GovCloud (US) the same day.

How much does Claude Fable 5.2 cost?

Anthropic lists it at $8 per million input tokens and $40 per million output tokens, with cache reads at $0.20 per million. Check the Amazon Bedrock pricing page for the exact Bedrock rate.

What is the Claude Fable 5.2 model ID on Bedrock?

The base ID is anthropic.claude-fable-5-2. You invoke it through us.anthropic.claude-fable-5-2 (Geo US) or global.anthropic.claude-fable-5-2 (Global).

What is the Claude Fable 5.2 context window?

1 million tokens, with up to 128K output tokens per request.

Claude Fable 5.2 vs Opus 5.5: which is better?

Fable 5.2 is Anthropic's more capable model for long, hard technical tasks, but it costs twice as much ($8/$40 vs $4/$20) and thinks harder by default. Opus 5.5 remains the better value for most coding and analysis.

Does Claude Fable 5.2 support zero data retention on Bedrock?

By default, no. It requires the aws_review retention mode, with up to 30 days of retention and human safety review. AWS says eligible customers can use it with zero data retention until Enterprise Frontier Safeguards is available.

Is Claude Fable 5.2 available in Europe on Bedrock?

Yes, but only through the Global inference profile, which can route requests worldwide. There's no EU Geo or in-Region endpoint listed.

What is Claude Mythos 5.2?

The same model without Fable 5.2's cyber and biology safeguards, available only to organisations verified through Anthropic's programmes. On Bedrock it's also offered to allow-listed customers in GovCloud (US).

Sources

• AWS model card: Claude Fable 5.2
• AWS What's New: Claude Fable 5.2 is now available on AWS
• AWS What's New: Claude Fable 5.2 in AWS GovCloud (US)
• AWS What's New: Claude Mythos 5.2 in AWS GovCloud (US)
• AWS Machine Learning Blog: Introducing Claude Fable 5.2 on AWS
• About Amazon: Claude Fable 5.2 is now available on AWS
• Anthropic: Claude Fable 5.2 model overview
• Anthropic: Claude Fable product page
• Amazon Bedrock pricing
• Hacker News launch discussion

Featured image: AWS data center buildings in Ashburn, Virginia (an AWS facility in the US East region, not a specific Fable 5.2 site). Photo by Vahurzpu, CC BY-SA 4.0, via Wikimedia Commons. Source