Claude Fable 5.1 · Anthropic · AI agents · Microsoft Foundry

Claude Fable 5.1: what Anthropic's new AI can do – and how companies can use it

Anthropic released Claude Fable 5.1 on 1 September 2026. The model is aimed at particularly demanding knowledge work, coding and long-running agentic tasks. What makes it especially interesting for companies is that Fable 5.1 can be used not only in the Claude chat, but also via Claude Code, enterprise surfaces, the API and as a preview in Microsoft Foundry. This article shows how to get started and what companies should look out for.

What is Claude Fable 5.1?

Claude Fable 5.1 is Anthropic's current Fable model for particularly demanding, long-running knowledge and coding tasks. Anthropic introduced it on 1 September 2026. It is designed to handle complex tasks with less ongoing supervision, use tools, detect errors and continue work across many steps.

A note on the name: as of September 2026, there is no official "Claude Fable 6". The current release is called Fable 5.1. Anthropic offers it as a generally available Fable model; alongside it, Mythos 5.1 is a more tightly restricted variant that is only available to vetted organisations, for example for cybersecurity and biology research.

Anthropic publishes the current product overview on the official Claude Fable page.

The difference from a classic chatbot: Fable 5.1 is particularly interesting when the goal is not a single answer, but a longer assignment involving research, files, tools, code or several intermediate steps.

The key features of Fable 5.1

1. Very large context window

Microsoft lists Fable 5.1 in Foundry with a context window of up to 1 million tokens and a maximum output of up to 128,000 tokens. This makes it possible to process large volumes of documents, extensive codebases or long project contexts within a single working context.

2. Adaptive thinking

The model supports adaptive reasoning. Put simply, it can apply more compute and reasoning effort to complex tasks and work more concisely on simpler ones.

3. Long-running tasks and agents

Anthropic explicitly positions Fable 5.1 for work that can run across many steps or even for hours. This includes coding projects, research, analysis and agentic workflows involving several tools.

4. Vision and document understanding

Fable 5.1 can analyse diagrams, tables and visual content in documents. This is relevant for reports, technical documentation, financial documents or architecture material, for example.

5. Self-verification

For complex tasks, the model can verify results more thoroughly, for example by testing its own code or checking intermediate results. This does not reduce the need for human oversight, but it does improve the workflow.

How do you use Claude Fable 5.1?

Option A: Claude on the web, desktop or mobile

For the simplest start, users open Claude and select Fable 5.1 in the model picker, provided their plan and access include the model. Anthropic lists Fable 5.1 for the paid Pro, Max, Team and Enterprise plans.

Option B: Claude Code

Developers can use Fable 5.1 in Claude Code. According to Anthropic, Fable 5.1 requires at least Claude Code version 2.1.255. The model is particularly useful for larger code changes, reviews, root cause analysis and tasks that span several files or repositories.

Option C: Claude for Microsoft 365 and other enterprise surfaces

Anthropic also lists Fable models for Claude for Microsoft 365, Claude Cowork, Claude Design and Claude Tag. Which surface makes sense in a company depends on whether individual knowledge workers, developers or automated workflows are to be supported. You can find the current access overview in the Claude Help Center.

Option D: API or Microsoft Foundry

For your own applications, Fable 5.1 can be integrated via the Claude Platform or supported cloud marketplaces. The model is also available in Microsoft Foundry – currently as a preview.

Briefing Fable 5.1 properly: define tasks instead of just prompting

With powerful agentic models, a long "mega prompt" does not automatically produce the best result. A clear task definition with a goal, material, boundaries and the desired outcome works better.

  • Goal: What should be available at the end – an analysis, code, a decision paper, a report or a prototype?
  • Context: Which files, systems or constraints does the model need to know about?
  • Quality criteria: How will you recognise that the task has been completed correctly?
  • Boundaries: Which systems may the model change, and which may it only read?
  • Verification: Should it run tests, check sources or explicitly flag assumptions?
  • Output format: For example a Markdown report, table, pull request, JSON or management summary.

A good brief is therefore more like "Analyse these three sets of documentation, compare the requirements with our current state, flag uncertainties and deliver a prioritised list of actions" than just "Summarise this".

Concrete use cases in companies

Software development

Analysing large codebases, implementing features, writing tests, investigating performance problems and preparing pull requests.

Research and knowledge work

Comparing many documents, bringing positions together, structuring sources and turning them into reports that support decisions.

Contracts and technical documentation

Searching document packages, extracting requirements and making differences between versions visible. For legal decisions, professional review naturally remains necessary.

Data and financial analysis

Interpreting tables, charts and reports, explaining anomalies and preparing management summaries.

Agentic workflows

Carrying out multi-step tasks across connected tools – for example gathering information, checking results, creating documents and deriving follow-up tasks. This is exactly where a model designed for long-running work plays to its strengths.

Claude Fable 5.1 in Microsoft Foundry

For Microsoft-oriented companies, the Foundry integration is particularly interesting. Microsoft currently lists claude-fable-5-1 as a preview, hosted on Anthropic infrastructure. There, Microsoft names a context window of 1 million tokens, a maximum of 128,000 output tokens, adaptive thinking, long-running tasks and tool use as core capabilities.

The architecture question is important: "available in Microsoft Foundry" does not automatically mean, in this preview, that the model itself is hosted on Azure infrastructure. Companies should therefore check data flows, region, contractual basis, logging and retention carefully before any production use.

Current model availability and technical properties are listed in the Microsoft Foundry documentation on Claude.

Costs, limits and model choice

Anthropic currently lists USD 10 per million input tokens and USD 50 per million output tokens for Fable 5.1. Cache reads are billed at a lower rate. In chat subscriptions, by contrast, the respective plan and usage limits apply instead of direct token billing.

For companies, model choice therefore matters. Fable 5.1 does not have to handle every email, short summary or simple text edit. High-performance models pay off especially where a task is complex, long, expensive in terms of human working time or spread across many documents and tools.

Sensible AI operations use different models depending on the task, quality requirements and budget.

Data protection, retention and security: check before the rollout

For Fable 5.1, Anthropic points out a standard 30-day data retention period for safety monitoring. Additional options or programmes exist for certain enterprise scenarios. These points can differ depending on the product surface, contract and cloud integration.

Before introducing the model, companies should therefore clarify at least the following:

  • Which data may be sent to the model?
  • Which personal, confidential or specially protected information is excluded?
  • Which retention and logging rules apply to the chosen access route?
  • Who may use Fable 5.1, and which tools may it be connected to?
  • How are results reviewed by subject-matter experts?
  • How are sensitive prompts, files and API keys protected?

Especially with agents that can access several systems, permission management is just as important as model quality.

When Fable 5.1 makes sense – and when a smaller model is enough

Fable 5.1 is above all interesting when a task involves a high proportion of complex reasoning, large amounts of context or many work steps. For short routine tasks, a faster and cheaper model can be more economical.

  • Fable 5.1: Large codebase, extensive research, complex analysis, long-running agent, demanding knowledge work.
  • Smaller/faster model: Short email, simple extraction, standard classification, minor text correction, frequent bulk task.
  • Human expert review: Legally binding decisions, security-critical changes, confidential approvals and results with high external impact.

The best AI strategy is not to always choose the most powerful model, but to find the right balance of quality, speed, cost and risk for each task.

Conclusion: Claude Fable 5.1 is particularly exciting for large, multi-step tasks

Claude Fable 5.1 is not "Fable 6", but Anthropic's current Fable version. Its strengths lie in complex knowledge work, coding, large amounts of context and long-running agentic tasks. Usage ranges from the regular Claude interface and Claude Code to API and Microsoft Foundry scenarios.

For companies, however, the technology alone is not enough. Access rights, data classification, costs, model choice, quality control and a well-run pilot determine whether a powerful model becomes a productive tool. büKOM Systemhaus GmbH supports you with AI strategy, Microsoft environments, secure integration and the development of sensible business workflows.

Why büKOM for Claude & business AI?

Because a powerful model only becomes truly valuable through a good business process

AI projects need more than model access: data, identities, the Microsoft environment, interfaces, security and clear use cases all have to work together.

Use case before tool We start with the work process and then choose the model, integration and level of automation.
Security built in Permissions, data flows, logging and approvals are considered from the outset.
Connecting Microsoft & AI Microsoft 365, Foundry, cloud and existing IT processes can be planned together.

You might also be interested in