Claude Fable 5.1: 1,000,000-Token Context, ¥68/¥340 Pricing, and Where It Fits

2026-11-06 · 原创·模型聚焦

Positioning: capability ceiling, broad availability

Per its official description, Claude Fable 5.1 is Anthropic's most capable widely available model, designed for hard reasoning, long-horizon agentic coding, multi-step research and complex knowledge work, with support for a 1,000,000-token context. Two phrases carry the weight: "most capable" sets the ceiling, "widely available" sets how you get it, so demanding production paths can be routed to it without a gated access process. Configure the public model ID `claude-fable-5-1`; the display name Claude Fable 5.1 is for presentation only. The modality is chat.

Specs and pricing

| Item | Value |

| --- | --- |

| Model ID | claude-fable-5-1 |

| Context window | 1,000,000 tokens |

| Input | ¥68 per million tokens |

| Output | ¥340 per million tokens |

| Modality | chat |

Input and output are billed separately, and output costs noticeably more. Two engineering implications follow. First, placing large material on the input side is the intuitive fit for long-context work. Second, output length must be constrained — fixed schemas, word caps, conclusions-only instructions — or the bill is driven by generation rather than reading. A 1,000,000-token window is not just "more room": it lets some pipelines skip the chunk-retrieve-reassemble step and send whole documents directly, removing a common failure mode where imperfect retrieval breaks the reasoning chain. Long context is still not free, so measure real token consumption on sample workloads before planning capacity.

Three scenarios that fit

Long-horizon agentic coding. Repository-level understanding, cross-file refactoring and iterative fixes require consistency across a long working session. A 1,000,000-token window keeps more code and interaction history in one place, so less information is lost to repeated summarization.

Multi-step research. Market scans, competitive analysis and technology selection demand synthesis across many sources plus layered reasoning. Passing source material directly into context yields more coherent output than stitching retrieved fragments together.

Complex knowledge work. Dense contract or technical-document analysis, long-form report drafting and consistency checks are input-heavy tasks with strict output requirements — a good fit for a high-capability model that produces a first draft for human review.

OpenAI-compatible request

curl https://平台 API 地址/v1/chat/completions \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-fable-5-1",
    "messages": [
      {"role": "user", "content": "Summarize the attached long document in under 300 words."}
    ]
  }'

The endpoint follows the OpenAI-compatible chat completions shape, so existing client code typically needs only a base URL and model ID change.

Choosing it

If your task is long-input, long-reasoning, and quality matters more than latency, Claude Fable 5.1 deserves the first test. For high-volume, short-text classification or extraction, a lighter model keeps costs predictable. Run a controlled comparison on real samples before splitting traffic, review the full spec and billing details in the model marketplace, and validate the integration with your own data.