Anthropic: Claude Fable 5.1

Claude Fable 5.1 is a text generation model from Anthropic, listed on api14 under the model id anthropic/claude-fable-5.1. It sits in the text category and is positioned by its provider as a broad improvement over Claude Fable 5, with the largest gains reported in agentic coding, long-running agentic workflows, and knowledge work. If you have used the earlier Fable release, this one is the direct successor to evaluate.

The model accepts text, images, and files as input, and it produces text as output. Its context window is 1,000,000 tokens, which allows very large inputs in a single request, such as long documents, sizeable codebases, or extended conversation histories. The maximum output is 128,000 tokens, so responses can be long as well, which matters for tasks like full file rewrites or detailed reports. Supported request parameters include max_tokens, max_completion_tokens, stop, response_format, structured_outputs, tool_choice, tools, reasoning, reasoning_effort, include_reasoning, and verbosity.

Those capabilities point to a few practical uses. Agentic coding is a natural fit: the long context and tool support let a model read across many files, call functions or external services, and carry out multi-step refactors without losing track of earlier context. Long-running agentic workflows benefit from the same combination of a large context window and tools, since intermediate results and prior steps can stay in the prompt. Because images and files are accepted as input alongside text, you can build document analysis, screenshot review, or mixed-content question answering on top of the same model. Knowledge work tasks such as summarising, extracting, and drafting also follow from the large input allowance and the structured output options.

To call it through api14, use the OpenAI-compatible chat completions interface with the model id anthropic/claude-fable-5.1. Because the API is OpenAI-compatible, existing OpenAI client libraries work once you point them at the api14 base URL and supply your key. You can pass the parameters listed above, so tool calling, structured outputs, response_format, stop sequences, and the reasoning controls (reasoning, reasoning_effort, include_reasoning) are all available in the request body, along with max_tokens or max_completion_tokens to bound the response and verbosity to adjust output length.

The main constraint to keep in mind is that output is text only; if your application needs generated images or audio, this model is not the right choice. The 128,000 token output ceiling is generous but still finite, so very large single-response jobs may need to be split. And if your task is small and simple, a lighter model may be sufficient; the value of the million-token context and agentic tooling shows up mainly on complex, long-horizon work.