Migration survival kit · revision 1.0
Claude Fable 5 → Fable 5.1 migration survival kit
Messages API integrations that manage tools, conversation history, retries, fallback routing, or prompt caching.
Deadline and scope
No Fable 5 retirement deadline is announced; complete these checks before routing traffic to Fable 5.1.
Documentation-derived guidance plus an executed Node.js request-shape preflight. No live Claude API inference was performed.
Prerequisites
- An inventory of every model ID and alias used by the integration
- Representative multi-turn request bodies with secrets and user data removed
- A workload-specific evaluation set for quality, cost, and latency
- A reversible traffic-control or model-routing plan
Five-minute preflight
- Change the model ID only in a controlled environment.
- Replace forced tool_choice values any or tool.
- Confirm thinking is not disabled or manually budgeted.
- Keep system, tools, and prior messages byte-stable across turns.
- Choose and monitor a thinking-block mismatch behavior.
- Exercise every route that can fall back to an older model.
- Re-baseline task quality, latency, token use, and cache behavior before cutover.
Before · Fable 5
{
"model": "claude-fable-5",
"tool_choice": {"type": "tool", "name": "record_summary"}
}After · Fable 5.1
{
"model": "claude-fable-5-1",
"tool_choice": {"type": "auto"},
"tools": [{"name": "record_summary", "strict": true}],
"messages": [{"role": "user", "content": "Call record_summary with the result."}]
}Compatibility matrix
| Check | Evidence class | Result |
|---|---|---|
| Model identifier | officially required | Change claude-fable-5 to claude-fable-5-1. Source |
| Forced tool choice | officially required | Replace any and named-tool forcing; both return 400 on Fable 5.1. Source |
| Conversation prefix | locally checked | The included preflight detects changed system, tools, or shared message prefixes in frozen fixtures. Source |
| Fallback continuity | recommended precaution | Benchmark the first older-model turn because Fable 5.1 thinking is removed on downgrade. Source |
| Cache economics | recommended precaution | Base token prices are unchanged; cache reads are documented at one quarter of the Fable 5 rate. Measure cost per completed task. Source |
Rollback conditions and procedure
- Prefer a conversation boundary for rollback rather than switching a live Fable 5.1 history to an older model.
- If an older model must receive the conversation, allow incompatible thinking blocks to be removed and treat the next turn as a re-planning event.
- Keep the previous request builder and model route available until the workload evaluation and fallback checks pass.
- Do not describe Fable 5 as retired: the official lifecycle table still lists it as active in this edition.
Known limits
- No claim is made about output quality, tool-call frequency, latency, or cost on a real workload.
- The local preflight checks request shape and prefix stability; it does not emulate Anthropic's API.
- Beta features and partner-platform behavior require verification in the operator's own environment.
Primary evidence
- Claude Fable 5.1 overviewAnthropic primary source · checked Sep 11, 2026
- Migrating to Claude Fable 5.1 and Claude Mythos 5.1Anthropic primary source · checked Sep 11, 2026
- Claude API errorsAnthropic primary source · checked Sep 11, 2026
- Claude Platform pricingAnthropic primary source · checked Sep 11, 2026
- Model deprecationsAnthropic primary source · checked Sep 11, 2026
Revision and corrections
- Initial pilot edition; no material correction recorded.