Last updated: 2026-09-07 — this guide now covers Claude Fable 5.1, released September 1, 2026 alongside Claude Mythos 5.1. Fable 5.1 introduces three breaking API changes versus Fable 5, so a migration that used to be a model-string swap now needs real code review. Everything below is current for 5.1; the original Fable 5 migration steps are preserved where they still apply.
Quick Answer
Migrating to Claude Fable 5.1 (model ID claude-fable-5-1) involves three mandatory fixes that did not exist in the Fable 5 migration:
- Forced tool calls now fail —
tool_choice: {"type": "any"}or{"type": "tool"}returns a 400 error. You must switch to automatic tool selection or prompt-based steering. - Thinking blocks are not backward-compatible — Fable 5.1's thinking blocks can't be passed back to older Claude models.
- Editing conversation history is restricted — modifying prior turns mid-conversation breaks on 5.1.
The upside: cache reads drop 75% to $0.25/M tokens (from $1.00), with the same $10/$50 list price as Fable 5. The simplest path for teams juggling multiple models remains routing through a gateway like TeamoRouter, where model switching is a single string change.
What Is Claude Fable 5.1?
Claude Fable 5.1 is Anthropic's current flagship, released September 1, 2026 together with Claude Mythos 5.1 — the same base model with stricter guardrails, available only through restricted ("trusted") access, while Fable 5.1 is generally available.
Key facts:
- Model ID:
claude-fable-5-1 - Context window: 1M tokens, 128K max output
- Pricing: $10 / $50 per 1M input / output tokens (unchanged from Fable 5), cache reads cut 75% to $0.25/M
- Estimated cost impact: Anthropic pegs typical workloads at ~25% cheaper than Fable 5, and up to ~45% cheaper for heavily agentic work — nearly all from the cache-read discount
- Available via the Anthropic API, and through gateways like TeamoRouter with a unified key
Breaking Change 1: Forced Tool Calls Return 400 Errors
The one that will break production code. On Fable 5 (and older models), you could force the model to call a specific tool:
# Fable 5 — worked
tool_choice = {"type": "tool", "name": "get_weather"}
# Fable 5.1 — returns a 400 error
Both {"type": "any"} and {"type": "tool", "name": ...} now fail. Anthropic's rationale: forcing a tool call skips extended thinking, which degrades argument quality (and closed a prompt-exploit path).
How to fix it:
# Fable 5.1 — let the model decide, steer with the prompt
tool_choice = {"type": "auto"}
# and in the system prompt: "Use the get_weather tool for any weather question."
Audit every call site that sets tool_choice. If you genuinely need deterministic invocation for a single step, make that call directly in your code rather than through the model.
Breaking Change 2: Thinking Blocks Don't Cross Model Generations
Fable 5.1's extended-thinking blocks cannot be read by earlier models. If you pass conversation history containing 5.1 thinking blocks back to Fable 5 or older, the request fails.
How to fix it: when routing a conversation to a different model generation, strip thinking blocks from prior turns first. Within the Fable 5.1 family (including Mythos 5.1) this is a non-issue.
Breaking Change 3: Don't Edit Conversation History
Modifying prior turns mid-conversation — rewriting an earlier user message, patching an assistant response to retry — is restricted on Fable 5.1.
How to fix it: restructure agents that "rewrite history" to instead append a corrective turn, or fork the conversation from the last clean state. This is the change behind the conversation history rewrite errors developers have reported since launch.
Migration Step 1: Update Model Identifiers
The mechanical part is still a model-string change:
| Model | Anthropic API Identifier | TeamoRouter Identifier |
|---|---|---|
| Claude Fable 5.1 | claude-fable-5-1 |
claude-fable-5-1 |
| Claude Fable 5 | previous Fable 5 ID | claude-fable-5 |
| Claude Mythos 5.1 | restricted access | n/a (trusted-access only) |
If you are using a gateway like TeamoRouter, the identifier follows the gateway's naming convention and routes to the correct upstream automatically.
Migration Step 2: Re-evaluate Cost Structure
Updated pricing for the current lineup (per 1M tokens):
| Model | Input | Output | Cache Read | Notes |
|---|---|---|---|---|
| Claude Fable 5.1 | $10 | $50 | $0.25 | ~25–45% cheaper than Fable 5 in practice |
| Claude Fable 5 | $10 | $50 | $1.00 | predecessor |
| Claude Opus 5 | — | — | — | see Anthropic pricing for current rates |
The cache-read cut is the real story: on agentic loops where cached system prompts and file context dominate input, moving from Fable 5 to 5.1 is a straight ~25–45% cost reduction with no code change beyond the model ID. See Fable 5.1 pricing explained for worked cost examples.
Migration Step 3: Testing and Validation
- Grep your codebase for
tool_choice— everyany/toolusage is a guaranteed 400 on 5.1. - Check history-handling code for thinking-block pass-through and prior-turn edits.
- Run your eval suite against 5.1: tool selection, long conversations, structured output.
- Canary 10% of traffic for a week, watching error rates (the 400s surface immediately) and cost (which should drop).
# Route 10% of traffic to Fable 5.1 for one week
if random.random() < 0.10:
model = "claude-fable-5-1"
else:
model = "claude-fable-5" # current model
With TeamoRouter, canary deployments are built-in: split traffic by percentage in the dashboard and monitor cost and error rates per model in real time.
Migration Checklist
- Replace every forced
tool_choice(any/tool) withauto+ prompt steering - Strip thinking blocks when crossing model generations in shared history
- Remove prior-turn editing from agent loops; append corrections instead
- Update model identifier strings to
claude-fable-5-1 - Recalculate cost projections with the $0.25 cache-read rate
- Run your test suite against Fable 5.1
- Implement canary deployment (10% traffic for one week)
- Compare costs and error rates, then switch fully
Using a Gateway to Simplify Model Migration
Model migration is one of the strongest use cases for an API gateway. With TeamoRouter:
- Model switching is a string change: no new SDKs, base URLs, or API keys.
- Canary deployments are built-in: split traffic by percentage and compare metrics side by side.
- Cost visibility per model: unified billing shows exactly what each model costs.
- No lock-in: if 5.1's stricter API rules break a workflow, fall back to Fable 5 instantly while you fix it.
FAQ
What are the breaking changes in Claude Fable 5.1?
Three: forced tool calls (tool_choice type any or tool) now return 400 errors; Fable 5.1 thinking blocks can't be read by older models; and editing prior conversation turns is restricted. All three have straightforward fixes covered above, per Anthropic's official migration guide.
What is the Fable 5.1 model ID?
claude-fable-5-1. Via a gateway like TeamoRouter you use the same string, and the gateway handles upstream routing.
Is Fable 5.1 more expensive than Fable 5?
No — list price is identical ($10/$50 per million tokens), and cache reads dropped 75% to $0.25/M. Anthropic estimates total cost down ~25% for typical workloads and up to ~45% for agentic ones.
Do I need to update the Anthropic SDK?
No SDK upgrade is required for the breaking changes; they're API behavior, not client version. Update your tool_choice usage and history handling in code.
Will my existing Claude Code setup work with Fable 5.1?
Yes — switch the model setting to Fable 5.1. If you route through TeamoRouter, you can change the model from the dashboard without touching Claude Code's configuration. Watch for agent flows that relied on forced tool calls or history rewriting; those patterns are what 5.1 rejects.
What is Mythos 5.1, and do I need to migrate to it?
Mythos 5.1 is the same base model as Fable 5.1 with stricter guardrails, available only through restricted trusted access. Most developers don't migrate to it — it exists for high-safety deployments. See Fable 5.1 vs Mythos 5.1 for the positioning.
Summary
The Fable 5 → 5.1 migration is no longer just a model-string swap: audit tool_choice, thinking-block pass-through, and history editing, then enjoy a 75% cache-read price cut on the same $10/$50 list price. Route the transition through TeamoRouter and you can canary, compare, and roll back without touching your SDK setup.
Get Started
Sign up for TeamoRouter and switch to claude-fable-5-1 from the dashboard — one key across Claude, GPT-6 Astra, and DeepSeek.