Claude Fable 5.1 is a proprietary language model from Anthropic, released in September 2026, for tasks that run for hours and integrate many tools: functions across an entire codebase, code review, multi-stage knowledge work, and agents that operate without continuous supervision. Fable 5.1 and Claude Mythos 5.1 are the same model with different levels of protective mechanisms. Fable 5.1 is generally available, Mythos 5.1 only through vetted access programs. The context window, i.e., the amount of text the model can consider in a request, holds one million tokens.
Crucial for enterprise use: Anthropic classifies Fable 5.1 as a Covered Model with mandatory data retention. Furthermore, Claude Opus 5.5, released in the same month, leads in all nine jointly published tests, is significantly cheaper per token, and available with Zero Data Retention. Read below for which use cases Fable 5.1 is still the right choice.
Claude Fable 5.1
Anthropic
September 2026
Proprietary (no open weights)
Proprietary, Mythos-class, Reasoning permanently active, Text and image input, Text output
Not yet released
Not yet released. Adaptive Reasoning, controllable via five effort levels
1,000,000 Tokens (max. 128,000 output tokens)
The generally available version with API identifier claude-fable-5-1. Usable via the Claude Platform, Amazon Bedrock, Google Cloud, and Microsoft Foundry, in Claude Pro, Max, Team, and Enterprise plans, as well as via our MaaS Proxy.
The same model with less restrictive protective mechanisms, designed for work in cybersecurity and biosciences. Mythos 5.1 is only available through Anthropic’s vetted access programs.
Claude Opus 5.5 was released in the same month, leads Fable 5.1 in all nine tests compared by Anthropic, and costs 60 percent less per token. Unlike Fable 5.1, it is not a Covered Model and is available with Zero Data Retention.
Within the Claude series, there are more cost-effective alternatives for many tasks. We will work with you to determine which model suits your requirements and what a change would specifically cost or save.
Anthropic publishes neither parameter count nor architecture and provides no information on Fable 5.1’s training data or training methods. The model’s knowledge cutoff is June 2026. It is confirmed that Fable 5.1 and Claude Mythos 5.1 are the same model and differ only in the scope of their protective mechanisms.
Reasoning, i.e., the step-by-step thinking through a task before responding, is permanently active and cannot be switched off. The depth of thought is controlled via five effort levels from low to max; on Amazon Bedrock, high is preset. This is the central cost lever: at low or medium, Fable 5.1 achieves results at the level of Fable 5 or better, according to Anthropic, at significantly lower costs.
Covered Model: Anthropic requires retention of requests and outputs for at least 30 days; Zero Data Retention only with explicit approval. With the announced Enterprise Frontier Safeguards, eligible customers should be able to keep data in their own cloud infrastructure. Until then, Anthropic assures these customers Zero Data Retention.
Protective mechanisms: Classifiers for cybersecurity and biosciences significantly increase the rejection rate. Penetration tests, exploit development, and research in biosciences are redirected by Anthropic to Opus models.
Behavior: Outputs carry watermarks according to the EU AI Act. In Gray Swan’s test, Fable 5.1, together with Opus 5.5, achieved the lowest success rate for prompt injection.
For tasks that run for a long time, integrate many tools, and whose results justify the higher price per token. For new projects without a dependency on Fable 5.1, Claude Opus 5.5 is generally cheaper and stronger in benchmarks. Fable 5.1 is also less suitable for mandatory Zero Data Retention or processing in Europe, for large volumes of standard tasks, and for regular questions about cybersecurity or biosciences. We would be happy to clarify this with you in an AI consultation.
Strong in agentic work: Significantly ahead of Opus 5 in Terminal-Bench-Science and AutomationBench.
Large context window: One million input tokens and 128,000 output tokens.
Effort levels as a cost lever: At low or medium, Fable 5.1 achieves results like Fable 5 at significantly lower costs.
Cost-effective cache reads: Cached request parts cost 75 percent less than with Fable 5.
Vision for documents: Captures diagrams, charts, and tables in files and PDFs.
Behind Opus 5.5: Claude Opus 5.5 leads in all nine jointly reported tests and is 60 percent cheaper per token.
Limited data protection: Minimum 30 days retention and no Europe-limited processing.
Weak in research requiring evidence: In Anthropic’s strict source check, Fable 5.1 did not pass any of the reports.
Increased rejection rate: Requests concerning cybersecurity and biosciences are more frequently rejected or redirected.
We compare Claude models for your specific use case and integrate them via our MaaS Proxy with central budget control and billing in Euros. Upon request, we can show you which applications you can operate in a GDPR-compliant manner with open models in our German data center.
Input and output cost the same per token as its predecessor Fable 5 and two and a half times as much as Claude Opus 5.5. Cached request parts are 75 percent cheaper than with Fable 5, and batch processing via the Anthropic API halves the base prices. Through our MaaS Proxy, you are billed in Euros.
Claude Opus 5.5 leads in all nine tests jointly reported by Anthropic and is 60 percent cheaper per token. More importantly for businesses: Opus 5.5 is available with Zero Data Retention and can be limited to European regions via Amazon Bedrock. As a Covered Model, Fable 5.1 requires a minimum of 30 days of retention and does not offer Europe-limited processing.
It is the same model. Mythos 5.1 has less restrictive protective mechanisms for cybersecurity and biosciences and is only available through Anthropic’s vetted access programs.
That depends on the use case. As a Covered Model, requests and outputs are retained for at least 30 days, and Europe-limited processing is not possible. Claude Opus 5.5 is available with Zero Data Retention and European processing. For data with high protection requirements, we will review the appropriate variant with you.
No. The model is proprietary and can only be used via Anthropic, connected cloud platforms, or an access point like our MaaS Proxy. Those who need full control over the infrastructure are better served with open models, such as GLM-5 or DeepSeek-V4-Pro.
Would you like individual advice?
Our AI experts are here for you!
