Claude Fable 5.1 Migration Guide: Model ID, Thinking and Client Setup

Use claude-fable-5-1 as the exact model ID. For Claude Code through LLMAPI, set the Anthropic base URL to https://llmapi.pro without /v1. Before switching a workflow, review its thinking configuration: Anthropic lists Fable 5.1 as using adaptive thinking that is always on.

Treat model selection, client configuration and account access as separate checks. Anthropic's documentation defines the model's capabilities; LLMAPI's catalog defines what the service lists for its plans. A catalog entry is not a successful request test for your account.

Choose Fable 5.1 for the right workload

The official model overview recommends Opus 5.5 as the starting point for most workloads. It positions Fable 5.1 for demanding reasoning and long-horizon agentic work, including cases where evaluations of Opus 5.5 at higher effort still fall short.

Use that distinction to frame your migration evaluation. Pick representative tasks from your own workflow and define what would justify switching before comparing results. For example, an illustrative evaluation could check whether a multi-step task reaches the required outcome and whether its tool interactions meet your application's requirements. These are suggested checks, not measured Fable 5.1 results.

Keep the version explicit. LLMAPI's subscription model list includes both claude-fable-5 and claude-fable-5-1 and instructs clients to use an exact model ID. If your intended destination is Fable 5.1, check for the full claude-fable-5-1 string rather than leaving the older ID in your configuration.

Review thinking settings before changing the model

Anthropic's overview lists Fable 5.1 with adaptive, always-on thinking. Do not plan a migration around turning thinking off.

For clients that discover capabilities programmatically, the official Models API documentation describes capabilities.thinking.types.disabled. This field reports whether a model accepts thinking: {type: "disabled"}. A supported value of false means that setting is rejected with a 400 error. This is an official Claude API capability description, not a claim that every gateway exposes the same discovery response.

Review your application and client settings for any explicit request to disable thinking. Make that a separate review item from the model ID: changing the name alone does not establish that the rest of your request configuration is appropriate.

For a controlled evaluation, begin with a fresh conversation and inspect the request settings you actually send. Treat tests involving tools or previously stored conversations as separate acceptance checks rather than assuming a model-selection test covers them.

Configure Claude Code's connection

The LLMAPI Claude Code guide documents the Anthropic protocol with a base URL that has no /v1 suffix. Its manual environment-variable setup is shown below with a placeholder key.

macOS or Linux:

export ANTHROPIC_BASE_URL=https://llmapi.pro
export ANTHROPIC_AUTH_TOKEN=YOUR_API_KEY

Windows PowerShell:

$env:ANTHROPIC_BASE_URL="https://llmapi.pro"
$env:ANTHROPIC_AUTH_TOKEN="YOUR_API_KEY"

Replace YOUR_API_KEY with your own key privately. The guide directs users to create keys in Console → API Keys.

For model selection, use the exact claude-fable-5-1 ID in your Claude Code or SDK model setting, as directed by the subscription catalog. The Claude Code guide's existing Fable examples use claude-fable-5; do not copy that older ID unchanged when your goal is Fable 5.1.

Do not mix protocol-specific base URLs. The LLMAPI documentation distinguishes the Anthropic base URL, https://llmapi.pro, from the OpenAI-compatible base URL, https://llmapi.pro/v1. Choose the URL for the protocol your client is using, not simply for the model's name.

Check account access and billing separately

LLMAPI lists claude-fable-5-1 among its subscription models. Its plans page identifies GET /v1/models on the gateway as the authoritative list of subscription-available models. Check that list with the account you intend to use before relying on the migration.

Subscription access and metered API billing are separate account checks. The plans page states that API billing uses a separate key and balance. If you are using metered billing, review the channel and pricing catalog rather than treating the subscription list as confirmation of your metered configuration.

For current subscription terms, consult the plans page. For metered prices, consult the channel catalog; it identifies the console ledger as authoritative. Neither catalog should be substituted for Anthropic's official model specifications.

Migration acceptance checklist

Before adopting the configuration for your workflow:

• Confirm that the intended model setting is exactly claude-fable-5-1.
• Match the base URL to the client protocol; Claude Code uses the Anthropic URL without /v1.
• Check the key and billing mode for the account you intend to use.
• Review any explicit thinking-disable setting against Fable 5.1's always-on thinking specification.
• Run a small fresh-conversation test before evaluating more complex workflows.
• Evaluate tool-dependent tasks and stored-conversation reuse separately, with acceptance criteria drawn from your application.

The configuration and catalog checks establish what to select and where to connect. Your own request and workflow tests should determine whether the migration is ready for your application.