Claude Fable 5.1 Long-Context Reasoning
Claude Fable 5.1 is Anthropic's top-tier long-context model for agentic coding and deep research. It can read images or PDFs and replies in text.
Overview
Claude Fable 5.1 is a Claude family model built by Anthropic. It applies always-on adaptive thinking to plan, reason, and write long, coherent answers. You can attach images or PDFs and ask questions about what’s inside. It returns a text response that you can paste into docs, tickets, or code reviews.
What you can build
- Long-horizon coding help for large repos, including refactors and multi-file changes
- Deep research drafts that keep multiple threads consistent across long outputs
- PDF analysis for reports, contracts, and technical docs, with cited snippets you can verify
- Diagram and chart explanations from screenshots, slides exported to PDF, or images
- Agent-style task plans with step-by-step execution notes and final deliverables
Inputs
- A required request written in plain text, describing the task and the desired format
- Optional attachments, limited to JPEG, PNG, GIF, WebP images, and PDF documents, with up to 50 files per run
- Optional conversation history provided as a JSON array of chat messages with roles and content; provide either a single prompt or a message list, not both
- Optional system-level instructions in plain text to set rules, tone, or constraints for the whole run
- Optional user and session identifiers (string or number) to group and continue chat history across runs
- Optional effort level selection (low, medium, high, xhigh, or max) to control how deeply the model thinks before answering
- Optional web search toggle to allow the model to look up up-to-date information during the run
- Optional response length cap from 1 to 128,000 tokens
- Optional tool behavior selection that supports only automatic tool calling or no tool calling
- Optional parallel tool calling toggle, if the run returns multiple tool calls in one turn
- Optional continuation inputs for tool-loop flows, including a prior task token and a list of tool results keyed by the model’s call IDs
Outputs
- A text response that may include headings, lists, tables, and fenced code blocks
- When tool calling is enabled and triggered, one or more tool call requests, each with a call ID and JSON arguments
- For continued tool-loop runs, a new task token that you can pass back with tool results to resume the same workflow
Limitations
- Claude Fable 5.1 supports text output only, even when you provide images or PDFs.
- You can’t force tool use. Only automatic tool calling or no tool calling is supported.
- Long agent sessions require stable conversation history. Editing earlier turns can break continuity.
- Safety classifiers can refuse some requests, especially in sensitive cybersecurity or biology areas.
- Low-quality inputs reduce accuracy. Scanned PDFs, blurry screenshots, and messy tables can cause missed details.
Safety & compliance
Claude Fable 5.1 applies production safeguards and may refuse disallowed requests. It can also flag or block some dual-use content, including parts of cybersecurity and biological research. Review outputs before you act on them, especially for security, legal, medical, or financial decisions. Follow your organization’s policies and applicable laws when you provide data and use results.
API quick start
Run fable 5.1 with a single API call.
{
"prompt": "Tell me about Türkiye",
"inputAll": "https://your-cdn.com/input.png",
"user_id": "...",
"session_id": "..."
}curl -X POST "https://api.wiro.ai/v1/Run/claude/fable-5-1" \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_WIRO_API_KEY" \
--data-binary @- <<'JSON'
{
"prompt": "Tell me about Türkiye",
"inputAll": "https://your-cdn.com/input.png",
"user_id": "...",
"session_id": "..."
}
JSON