Claude Fable 5.1 in Europe: Price and Retention

TL;DR

Claude Fable 5.1 is Anthropic’s frontier model for long-horizon agents. In Europe, only Google Cloud’s EU multi-region keeps inference in EU member states. Bedrock has no EU geo ID. As a Covered Model it retains prompts for 30 days everywhere. Start with Opus 5 unless your evals fail.

What is Claude Fable 5.1 and when should you use it?

Claude Fable 5.1 is Anthropic’s current frontier model for demanding reasoning and long-horizon agent work. Anthropic released it on 1 September 2026, with a 1 million token context window, 128,000 token max output, and always-on adaptive thinking at default effort high. For most company workloads we still start on Opus 5 and promote only when evals stall.

The Fable 5.1 overview is explicit about that split. Use Fable 5.1 for multi-hour agent harnesses, whole-codebase migrations, and dense document or spreadsheet work. Use Opus 5 as the default. Claude Mythos 5.1 shares the same capabilities and price, but only Project Glasswing participants get it. The knowledge cutoff is June 2026. Anthropic will not retire Fable 5.1 before 1 September 2027. Google Cloud’s own date is earlier: not sooner than 1 March 2027.

Adaptive thinking cannot be turned off. You steer depth with effort: low, medium, high, xhigh, max. Temperature must be 1.0 or unset. Forced tool choice (tool_choice of any or a named tool) returns a 400 error. Leave it at auto and set strict: true on the tool schema.

The current tokenizer, introduced with Claude Opus 4.7, produces about 30% more tokens for the same text. Anthropic says 1 million tokens is roughly 555,000 words. A budget copied from a Sonnet 4.x bill will undershoot. Token prices for the rest of the lineup sit in our Claude API pricing EU comparison.

Where can you run Claude Fable 5.1 with EU processing?

Google Cloud’s Europe multi-region is the only managed Claude Fable 5.1 route that keeps ML processing in EU member states. Amazon Bedrock publishes a US geo ID and a global ID, not an eu. profile. Microsoft Foundry offers Fable only as Hosted on Anthropic, Global Standard, so data can leave Azure and leave your selected region.

We read the three platform pages on 1 October 2026.

Platform	Model ID	EU processing	Retention you cannot skip
Claude API	claude-fable-5-1	No EU inference geo	30 days at Anthropic
Amazon Bedrock	global.anthropic.claude-fable-5-1 or us.anthropic.claude-fable-5-1	No. Runtime geo ID is US only	aws_review, up to 30 days inside AWS
Google Cloud	claude-fable-5-1	Yes, Europe multi-region	Up to 30 days in Google Cloud
Microsoft Foundry	claude-fable-5-1	No EU data zone	Anthropic-hosted, 30 days

The Bedrock model card lists us.anthropic.claude-fable-5-1 as the only geo inference ID and global.anthropic.claude-fable-5-1 as the global ID. In-region on bedrock-runtime is N/A. AWS sample code calls the global ID from us-east-1. Copying a Sonnet 5 pattern such as eu.anthropic.claude-fable-5-1 will fail because that ID does not exist. Our Claude on AWS Bedrock EU guide covers the models that do have an EU profile.

Google’s Fable 5.1 page lists model availability and ML processing in the Europe multi-region, plus the United States multi-region and the global endpoint. That is the EU option. The endpoint, IAM roles and eu region string are in our Claude on Vertex AI in Europe guide.

Foundry is the wrong pick for EU residency. Microsoft’s hosting comparison puts Fable and Mythos on Anthropic infrastructure only. Data might be processed outside Azure, including outside the selected Azure region. Hosted on Azure covers Opus, Sonnet and Haiku, not Fable. Anthropic remains the processor on both Foundry hosting options.

What does the 30-day Covered Model retention mean?

Claude Fable 5.1 is a Covered Model. Anthropic requires 30 days of prompt and output retention on every platform where the model is offered. Zero data retention is not available unless Anthropic expressly authorizes it. On the Claude API, a workspace without that setting gets a 400 invalid_request_error.

The API retention page and the Covered Models article agree on the rule. The policy took effect on 9 June 2026. Fable 5 and 5.1 share the underlying model with Mythos 5 and 5.1, with extra safeguards in cyber and bio. Anthropic keeps the logs so classifiers can see patterns across many requests, not one prompt at a time.

Where the data sits depends on the platform:

• Claude API and Claude Platform on AWS. Anthropic holds the retained data. ZDR organizations can turn on 30-day retention for one workspace and leave the others on zero retention.
• Amazon Bedrock. You must set data retention to aws_review. Retained data stays in AWS. AWS may review it. The model card says the provider does not receive the content.
• Google Cloud. You must accept the Advanced AI Safety Addendum. Google stores prompts and responses for up to 30 days for abuse monitoring. Retained data stays in Google Cloud.
• Microsoft Foundry. Retention is per Azure subscription. A subscription already on ZDR needs a separate subscription to call Covered Models.

Do not send special-category data under GDPR Article 9 into Fable 5.1 unless your DPO has signed off on a 30-day safety log. For that class of prompt, stay on Opus 5 or Sonnet 5 with zero retention, or self-host an open-weight model. The wider provider matrix is in our Claude GDPR comparison.

What does Claude Fable 5.1 cost in 2026?

Claude Fable 5.1 costs 10 USD input and 50 USD output per million tokens on the Anthropic API and on Google’s global endpoint. The Google Cloud EU multi-region adds the published 10% regional premium: 11 USD input and 55 USD output. Batch halves both sides. Cache reads on Fable 5.1 cost 0.25 USD per million tokens, a quarter of the usual 10% cache-read rate.

All figures below are USD per million tokens, read on 1 October 2026 from Anthropic pricing and Google Cloud Agent Platform pricing. No vendor publishes euro list prices.

Route	Input	Output	5 min cache write	Cache read	Batch in / out
Claude API (global)	10.00	50.00	12.50	0.25	5.00 / 25.00
Google Cloud global	10.00	50.00	12.50	0.25	5.00 / 25.00
Google Cloud EU multi-region	11.00	55.00	13.75	0.275	5.50 / 27.50
Bedrock EU geo	not offered	not offered	not offered	not offered	not offered

A one-hour cache write is 20.00 USD globally and 22.00 USD on Google EU. Anthropic prices cache reads at 0.025 times the base input price on Fable 5.1, versus 0.1 times on most other Claude models. That is the one place Fable is cheap relative to its list price. Long agent loops that re-read a cached prefix pay for the model through output tokens, not through those cache hits.

Opus 5 remains 5 / 25 USD globally and 5.50 / 27.50 USD on Google EU and Bedrock EU. If a task finishes on Opus 5, Fable 5.1 costs twice as much on input and twice as much on output. We recommend a written promotion rule: run the eval pack on Opus 5 first, then on Fable 5.1 at medium and high, and keep Fable only where the quality gap pays for the tokens.

Foundry and Claude Platform on AWS bill in Claude Consumption Units at 0.01 USD per CCU. One hundred CCU equals 1.00 USD of Anthropic list fees. That is a billing unit, not a discount.

How do you call Claude Fable 5.1 on Google Cloud in the EU?

Call Claude Fable 5.1 from the Europe multi-region with model ID claude-fable-5-1 after you accept the Advanced AI Safety Addendum on the Model Garden card. Set the Anthropic Vertex client region to eu. Do not point the same app at Bedrock with an invented eu. prefix. Confirm spend on the EU price row, 11 / 55 USD, before you open the model to production traffic.

1. In a Google Cloud project with billing, enable the Gemini Enterprise Agent Platform API (aiplatform.googleapis.com).
2. Open the Claude Fable 5.1 model card, accept the Advanced AI Safety Addendum, and enable the model for the Europe multi-region.
3. Grant roles/aiplatform.user to the runtime identity. Keep the project off the global endpoint if residency is the reason you chose Google Cloud.
4. Install the Vertex extra (pip install -U 'anthropic[vertex]') and authenticate with application default credentials.
5. Pin the model ID. Do not use a generic fable alias in production.

import os
from anthropic import AnthropicVertex

client = AnthropicVertex(
    project_id=os.environ["GOOGLE_CLOUD_PROJECT"],
    region="eu",
)

message = client.messages.create(
    model="claude-fable-5-1",
    max_tokens=1024,
    messages=[
        {
            "role": "user",
            "content": "Summarise this filing in six bullets, then list open risks.",
        }
    ],
)
print(message.content)

Google’s own request guide shows anthropic[vertex] and the model name claude-fable-5-1. The sample on that page uses a US regional host. For EU processing, region="eu" is the string our Vertex guide already documents for the Europe multi-region. Handle stop_reason: "refusal" as a normal path. Prompt-stage refusals are not billed. Mid-stream refusals are billed for tokens already produced.

If you only have AWS today and the prompt cannot leave the EU, do not force Fable 5.1 onto Bedrock. Keep that workload on eu.anthropic.claude-opus-5 or eu.anthropic.claude-sonnet-5, or stand up a Google Cloud project for the Fable jobs alone.

FAQ

How much does Claude Fable 5.1 cost for an EU company?

On the Anthropic API and Google’s global endpoint, 10 USD input and 50 USD output per million tokens. On Google Cloud’s EU multi-region, 11 USD and 55 USD. Batch is half of those pairs. A cache read is 0.25 USD globally and 0.275 USD on Google EU. Bedrock has no EU geo price because it has no EU geo ID.

Claude Fable 5.1 vs Claude Opus 5: which should we default to?

Default to Opus 5. Anthropic says to start there for most workloads and to move to Fable 5.1 when Opus 5 at higher effort still falls short. Opus 5 is 5 / 25 USD globally and has an EU geo profile on Bedrock. Fable 5.1 is twice that list price, has no Bedrock EU geo ID, and forces 30-day retention.

Does Amazon Bedrock keep Claude Fable 5.1 in the EU?

No. The Bedrock model card publishes us.anthropic.claude-fable-5-1 and global.anthropic.claude-fable-5-1. There is no eu.anthropic.claude-fable-5-1. The global ID can route to any commercial AWS region. You also have to enable aws_review.

Can we use Claude Fable 5.1 with zero data retention?

Not unless Anthropic expressly authorizes it. Covered Models require 30-day retention on every platform. On the Claude API the request fails with a 400 error if the workspace has no retention. On Bedrock you must set aws_review. On Google Cloud you accept the Advanced AI Safety Addendum.

Is Claude Fable 5.1 available on Microsoft Foundry in Europe?

The model ID claude-fable-5-1 is in Foundry as a preview, Hosted on Anthropic, Global Standard only. Microsoft says data might be processed outside Azure, including outside your selected region. There is no Hosted on Azure option and no EU data zone for Fable.

When does Claude Fable 5.1 retire?

Anthropic: not sooner than 1 September 2027. AWS matches that date. Google Cloud: not sooner than 1 March 2027. If you ship a product on Google Cloud, plan a model migration before Google’s date, not Anthropic’s.

Sources

1. Anthropic docs: Claude Fable 5.1 overview (1 October 2026)
2. Anthropic docs: Pricing (1 October 2026)
3. Anthropic docs: API and data retention (1 October 2026)
4. Anthropic Privacy Center: Covered Models (1 October 2026)
5. AWS Bedrock: Claude Fable 5.1 model card (1 October 2026)
6. Google Cloud: Claude Fable 5.1 (1 October 2026)
7. Google Cloud: Agent Platform pricing (1 October 2026)
8. Microsoft Learn: Claude hosting comparison (1 October 2026)