Claude Fable is a family of large language models developed by Anthropic for software engineering, research, document analysis, and extended, multistep tasks. It combines text and image understanding with tool use in AI agent systems. Introduced with Claude Fable 5 on June 9, 2026, the family provides broadly available access to capabilities associated with Claude Mythos, but with additional safeguards restricting certain uses. As of October 15, 2026, its latest released version is Claude Fable 5.1, launched on September 1, 2026. (anthropic.com)
Development and releases
Anthropic introduced Fable 5 alongside Mythos 5. The company described them as the same underlying model deployed with different safeguards: Fable was intended for general availability, while Mythos initially served a limited group of cybersecurity defenders and infrastructure providers. Fable’s restrictions could redirect some requests to a less capable Claude model rather than permit the underlying model to answer them directly. (anthropic.com)
Access to both models was suspended on June 12, 2026. According to Anthropic’s redeployment announcement, immediately effective United States export controls required restrictions on access by foreign nationals, and the company suspended all access because it lacked a reliable means of verifying nationality in real time. Anthropic announced that the controls had been lifted on June 30; access was restored on July 1. (anthropic.com)
Fable 5.1 followed on September 1, 2026. Its release emphasized improvements in long-running coding, multistep research, and work involving documents, spreadsheets, and presentations. It retained Fable 5’s standard input and output token prices while lowering the price of cached-input reads. (platform.claude.com)
Capabilities and uses
Fable is designed for work extending beyond a single conversational response. In an agent environment, it can plan tasks, call tools, inspect intermediate results, and continue working across multiple applications. Anthropic identifies enterprise analysis, codebase-wide software changes, and extended research as principal applications. These capabilities depend partly on the tools and execution environment supplied around the model. (anthropic.com)
In software engineering, its documented uses include implementing features across a codebase, reviewing code, improving performance, writing tests, and checking visual outputs against a design. Its image understanding also supports analysis of charts, diagrams, and tables in document-heavy workflows. Anthropic presents these as intended capabilities rather than guarantees that every project will be completed correctly or autonomously. (anthropic.com)
Fable 5.1 accepts text and images and produces text. Its context window accommodates up to one million tokens, with a maximum output of 128,000 tokens. Its documented knowledge cutoff is June 2026. The model uses adaptive thinking, which is always enabled; configurable effort settings govern the depth of its reasoning process. (docs.aws.amazon.com)
Relationship to Claude Mythos
The distinction between Fable and Mythos concerns deployment safeguards and access, rather than an entirely separate underlying model. Anthropic states that Fable 5.1 and Mythos 5.1 share the same underlying model, while Mythos is available through verification programs supporting activities such as advanced cybersecurity and life-sciences research. (anthropic.com)
This distinction affects observed performance. A task that the underlying model can perform may nevertheless be blocked or routed elsewhere in the Fable deployment. In its published evaluations, Anthropic reports that safeguard interventions affected some benchmark results, including cases where another Claude model completed a restricted task. Consequently, reported scores describe a particular deployment and evaluation configuration, not simply unrestricted model capability. (anthropic.com)
Access and pricing
As of October 15, 2026, Fable 5.1 is offered through Anthropic’s application programming interface and several cloud computing platforms, including Amazon Bedrock, Google Cloud, and Microsoft Foundry. Its Claude API identifier is claude-fable-5-1. Standard pricing is 10 US dollars per million input tokens and 50 US dollars per million output tokens; cached-input reads cost 0.25 US dollars per million tokens. Subscription access and API billing are distinct arrangements. (platform.claude.com)
Safeguards and operational limitations
Fable 5.1 includes classifiers restricting some dual-use cybersecurity and life-sciences content. Amazon Bedrock documentation notes that its refusal rate is materially higher than that of earlier Claude models. A blocked request can return a normal HTTP response containing a refusal stop reason, making refusal handling a separate application-level concern from network or server failure. (docs.aws.amazon.com)
Safeguards can also block benign work. Anthropic reported reducing cybersecurity false positives in the 5.1 release and allowing vulnerability discovery while retaining restrictions on exploit development. Its published results therefore distinguish between improved underlying capabilities and the practical boundaries imposed by deployment controls. (anthropic.com)
References
- Claude Fable 5 and Claude Mythos 5anthropic.com
- Redeploying Fable 5anthropic.com
- Introducing Claude Fable 5.1 and Claude Mythos 5.1anthropic.com
- Claude Fable 5.1 - Claude Platform Docsplatform.claude.com
- Claude Fableanthropic.com
- Claude Fable 5.1 - Amazon Bedrockdocs.aws.amazon.com