What is Claude Fable 5?
Claude Fable 5 is an active frontier model from Anthropic. It is aimed at demanding workloads that may involve sustained reasoning, multi-step planning, tool calls, code generation, document analysis, and autonomous execution over extended periods. Anthropic describes the model as suitable for tasks that can run for hours or days, rather than only short conversational exchanges.
In practical terms, Fable 5 is intended for work such as coordinating a complex research process, developing or reviewing a substantial software project, analyzing a large collection of documents, or operating an agent that must complete several dependent steps. The model can plan across stages, use tools, delegate work to sub-agents, and check its results according to the supplied research.
The model launched on June 9, 2026. Its documented knowledge cutoff is January 2026, so information learned after that date requires retrieval or another current data source. Web search is supported through compatible Anthropic API workflows, but using a search tool does not change the model’s underlying knowledge cutoff.
Where Fable 5 fits in Anthropic’s lineup
Fable 5 occupies Anthropic’s frontier-model tier for difficult reasoning and long-running agentic work. The supplied documentation does not present it as a general-purpose low-cost or low-latency default. Instead, its positioning reflects a capability-first trade-off: developers can use it when the value of deeper reasoning, larger context, and more reliable multi-stage execution justifies higher token costs and slower responses.
Anthropic also documents fallback behavior involving Claude Opus 4.8 for some blocked cybersecurity and biology requests. That behavior is part of supported safety workflows, not evidence that Fable 5 and Opus 4.8 have identical capabilities or pricing. Developers should treat refusals and fallback responses as normal outcomes that their applications need to handle.
Core specifications at a glance
| Specification | Claude Fable 5 |
|---|---|
| Provider | Anthropic |
| Release date | June 9, 2026 |
| Status | Active; tentative retirement is not sooner than June 9, 2027 |
| Model ID | claude-fable-5 |
| Context window | 1,000,000 tokens |
| Maximum output | 128,000 tokens per request |
| Knowledge cutoff | January 2026 |
| Input | Text and images |
| Output | Text |
| Reasoning | Adaptive thinking with configurable effort |
A token is a unit of text used for billing and context processing; it may represent a whole word, part of a word, punctuation, or other text. The 1-million-token context window is the total working space available for the request, including instructions, conversation history, documents, tool content, and the generated response. The 128,000-token output limit is separate: it describes the maximum response size supported for a request.
Reasoning, coding, and long-horizon work
Fable 5’s main distinction is its intended use for problems that require several connected decisions. Adaptive thinking is always enabled, while the effort level can be configured. This gives developers a way to trade response depth against resource use for different tasks, although the supplied research does not define a universal latency or quality result for each effort setting.
The model is particularly suited to advanced coding projects. Useful examples include reasoning through a multi-file change, planning an implementation before writing code, investigating a difficult defect, reviewing a large codebase, or coordinating tool-assisted development steps. Its value in these cases comes from maintaining more context and following a longer plan, not simply from producing a short code snippet.
Anthropic also positions Fable 5 for multi-stage research and complex knowledge work. A workflow might involve examining uploaded documents, identifying gaps, searching for current information, comparing evidence, and producing a structured report. Developers should still validate important facts and code: a large context window and deeper reasoning do not guarantee factual accuracy or correct execution.
Supported inputs, outputs, and tools
Claude Fable 5 accepts text and image input and produces text output only. It can therefore interpret visual material supplied in a request, but it does not natively generate images, audio, or video. This distinction matters for product design: an application can use Fable 5 to analyze an image or create a textual description, but a separate generation system is required for direct media creation.
The model supports tool use, streaming responses, prompt caching, structured outputs, and batch processing. Tool use allows an application to expose functions or external capabilities that the model can call during a task. Streaming can begin delivering a response before the complete result is ready. Structured outputs help applications request machine-readable results that conform to an expected structure; this should not automatically be treated as a separate, provider-defined JSON mode.
Web search is available through supported Anthropic API deployments and related platforms. Search can help with information newer than the January 2026 cutoff, but developers remain responsible for handling citations, source quality, failures, and conflicting results.
Availability and deployment options
According to the supplied documentation, Claude Fable 5 is available through the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS. Availability and supported features can differ by platform, so the model name alone does not guarantee that every tool, caching option, safety behavior, or output feature is exposed identically in every deployment.
The canonical Anthropic API model ID is claude-fable-5. Teams selecting a cloud-hosted deployment should verify the platform-specific model identifier, regional availability, request limits, billing rules, and supported features before committing an application architecture.
Pricing and cost trade-offs
Anthropic lists standard Claude Fable 5 pricing at $10 per million input tokens and $50 per million output tokens. Input tokens include the content sent to the model, while output tokens cover the generated response. Long prompts, large documents, tool results, and extended conversations can therefore increase input usage even when the visible answer is relatively short.
Prompt caching is priced separately. Cache writes cost $12.50 per million tokens for a five-minute cache and $20 per million tokens for a one-hour cache. Cache reads cost $1 per million tokens. Caching may be useful when the same large instructions or reference material are reused across requests, but the application must account for cache-write charges and the selected cache duration.
Batch processing applies a 50% discount to standard input and output token prices. That produces effective batch prices of $5 per million input tokens and $25 per million output tokens. Batch processing is most relevant when work does not need to complete interactively; it is less appropriate for a user waiting for an immediate answer.
These prices make Fable 5 a poor fit for routine, high-volume generation when a less expensive model can achieve the required result. Its economics are easier to justify when reducing human review, completing difficult coding work, processing large information sets, or successfully finishing a long-running task is more valuable than minimizing every token.
Strengths and limitations
Where Claude Fable 5 is strongest
- Long-horizon reasoning that requires planning and several dependent steps.
- Advanced coding, codebase analysis, debugging, and tool-assisted development.
- Large document and knowledge-work workflows supported by a 1-million-token context window.
- Multi-stage research that combines documents, tools, search, and synthesis.
- Agentic tasks involving tool calls, delegation, verification, and extended execution.
- Applications that benefit from configurable reasoning effort, structured outputs, streaming, caching, or batch processing.
Important limitations
- It is text-output oriented and does not natively generate images, audio, or video.
- Its $10 input and $50 output per-million-token standard pricing is relatively high for simple or high-volume tasks.
- It is not positioned as the fastest option; the supplied editorial assessment rates its speed below its reasoning and coding capabilities. That rating is an editorial estimate, not an Anthropic-published benchmark.
- A large context window does not eliminate the need to manage prompt size, document relevance, tool results, and application costs.
- Safety classifiers may refuse some potentially harmful cybersecurity and biology requests.
- The January 2026 knowledge cutoff means current facts need retrieval or verification.
- Platform availability and feature support vary across Anthropic API, cloud, and partner deployments.
The supplied editorial scores rate Fable 5’s reasoning at 10 out of 10, coding at 9 out of 10, speed at 3 out of 10, and cost at 4 out of 10. These are comparative editorial judgments rather than provider-published measurements, and they should not be read as standardized benchmark results.
When to choose Claude Fable 5
Choose Claude Fable 5 when the task is difficult enough that planning quality, context capacity, tool coordination, or completion reliability matters more than minimum cost and response speed. Strong candidates include autonomous software agents, complex code migrations, deep document analysis, long-running research, technical investigations, and enterprise workflows where a failed intermediate step can be expensive.
Another option may be more appropriate for short answers, simple classification, routine extraction, inexpensive content generation, or latency-sensitive interactions. A smaller or faster model can be a better default when the task has limited context and a straightforward success criterion. A dedicated image, audio, or video model is necessary when the required result is native media rather than text.
Fable 5 is also not automatically the right choice simply because a request contains many documents. If the documents are repetitive, the task is highly standardized, or the response can be generated with a short deterministic workflow, caching or a less expensive model may provide better economics. The strongest case for Fable 5 is a problem where sustained reasoning and multi-step execution materially improve the chance of completing the work correctly.
Lifecycle and safety considerations
Anthropic currently lists Claude Fable 5 as active, with a tentative retirement date not sooner than June 9, 2027. Because the retirement date is described as tentative or not sooner than that date, production users should still monitor Anthropic’s lifecycle documentation rather than treat it as an unconditional guarantee.
Applications should also be designed for refusals, tool failures, incomplete results, and fallback behavior. Safety restrictions are part of the model’s operating environment, particularly for certain cybersecurity and biology requests. A robust integration should communicate such outcomes clearly, avoid assuming every request will produce ordinary text, and preserve human review for high-impact decisions.