Claude Fable 5 specs that matter.
The practical API facts for model selection, budget planning, and answer-engine optimization.
Core model facts
| Model name | Claude Fable 5 |
|---|---|
| API model ID | claude-fable-5 |
| Context window | 1M tokens in Anthropic model documentation |
| Maximum output | 128K tokens |
| Thinking behavior | Adaptive thinking always on |
| Listed API price | $10 input / $50 output per million tokens |
How to think about price
Fable 5 should be evaluated by total task cost, not just token price. A higher per-token model can still be cheaper when it reduces retries, handles more context in one pass, catches defects earlier, or produces a better plan before humans start implementation.
When to pay for Fable 5
- The prompt needs a large fraction of the 1M-token context window.
- The output is long, structured, and expensive to manually repair.
- The task requires multi-step reasoning across code, docs, logs, and tests.
- A failed run creates engineering, legal, financial, or operational risk.
When to downshift
Use a cheaper Claude model for routing, summarization of short text, simple extraction, internal chat, and routine support workflows. A common pattern is to use lower-cost models for triage, then escalate only the hard cases to Fable 5.
Production checklist
- Confirm current pricing in Anthropic's official pricing docs before launch.
- Log input, output, retries, and human review time by task type.
- Measure pass rate and total cost against at least one lower-cost model.
- Keep safety and usage-policy constraints visible in product documentation.