Identity and current terms
| Provider / API ID | Anthropic / claude-fable-5-1 |
|---|---|
| Context / maximum output | 1,000,000 / 128,000 tokens |
| Thinking | Adaptive and always on; default effort high |
| Base input / output | $10.00 / $50.00 per 1M tokens |
| Cache | $12.50 five-minute write; $20 one-hour write; $0.25 hits/refreshes per 1M |
| Versioning | Provider documents claude-fable-5-1 as a fixed dateless snapshot |
Check compatibility before ranking
Anthropic documents adaptive thinking as always on. Forced tool choice returns an error; strict automatic tool use or structured output is the documented alternative. Earlier thinking blocks can become invalid when conversation history is edited, and fallback can change which model answers.
Confirm the account’s retention authorization before sending repository data. A model that cannot meet tool-schema or data-policy requirements is not a viable upgrade for that workload, regardless of headline capability.
Cost is more than the headline pair
An illustrative 80K uncached input plus 10K billed output tokens costs $1.30. Repeated work can have different economics because Fable 5.1’s cache hits are inexpensive while writes depend on the selected lifetime. Cache expiry, edited prefixes, thinking tokens, tools, retries, and human review remain material.
Evidence limits
External coding reports are useful for discovery, but results from a complete product or harness cannot be cleanly attributed to model weights alone. This research did not establish same-task, same-harness accepted outcomes for Fable 5.1 against Luna, Terra, or Astra.
Compare its token economics in the workflow-cost analysis or return to the buyer guide.
Reader questions
Frequently asked questions
What does Claude Fable 5.1 cost?
The direct Anthropic rates checked on 12 September 2026 are $10 per million base input tokens and $50 per million output tokens, with separate cache-write and cache-hit prices.
Can I force a tool call with Fable 5.1?
The provider documentation says forced tool use returns an error. It recommends automatic strict tool use or structured outputs, plus prompting that states when the tool applies.
Primary reading
Sources and update conditions
Rates, availability, model identity, and provider behavior can change. Recheck the linked sources before making a consequential purchase or sending repository data.