Claude Fable 5.1 in Claude Code: Upgrade and Verify Safely

To use Claude Fable 5.1 in Claude Code, update the client, start a fresh session, open /model, and select Fable 5.1. Then run /status and verify that the active model is Fable 5.1 before giving it a real task. If the model is missing, do not guess a model identifier or enable paid usage just to test it. Confirm the account, plan, organization policy, and client update path first.

Anthropic released Fable 5.1 on September 1, 2026. The model is generally available across Anthropic's platforms, while Mythos 5.1 is limited to trusted-access programs. Anthropic says Fable 5.1 defaults to High effort in Claude Code, so a successful model selection can also change task cost and latency. Treat the upgrade as a controlled change, not a reason to start an unattended write immediately.

What Changed in Fable 5.1

Fable 5.1 is a new model version, not a silent patch to Fable 5. Anthropic's launch notes identify four practical changes:

• stronger performance on its published coding and long-running-task evaluations;

• similar or better results than Fable 5 at Low or Medium effort in Anthropic's tests;

• cache reads priced at $0.25 per million tokens for usage billed by token, 75% below Fable 5's cache-read price; and

• more precise safeguards, including fewer false positives in Anthropic's cybersecurity evaluation.

Those are provider-reported results, not a guarantee for every repository. Anthropic's cost estimate depends on workload shape and cache reuse. Your first goal is simpler: prove that Claude Code selected the intended model, can read the current repository, and respects its existing approval boundary.

Step 1: Record the Current Working State

Before updating, finish or stop the current turn and preserve any useful output. In an ordinary shell, record the installed version:

claude --version

Inside Claude Code, run:

/status

Record the active model, account or organization, and loaded settings sources. Do not post a complete status dump publicly because it can reveal local paths, organization details, and configuration.

If you are in the middle of a repository change, inspect git status separately and save the relevant work. A model update should not be combined with a branch reset, dependency upgrade, or permission change.

Step 2: Update Through the Owning Installer

For Anthropic's native Claude Code installation, run:

claude update
claude --version

If Claude Code reports that another package manager owns the installation, use that manager's documented update path. Do not install a second copy over the first. After updating, fully close old Claude Code processes and start a fresh terminal session so the new executable and model catalog are loaded together.

Anthropic's public CLI reference documents claude update, but its launch post does not name one universal minimum Claude Code version for Fable 5.1. Verify availability in the live model picker instead of copying a community-reported version threshold.

Step 3: Select Fable 5.1 From the Model Picker

Start Claude Code in the intended repository and run:

/model

Choose the visible Fable 5.1 entry. Then run:

/status

Expected result: the active model shown by /status is Fable 5.1, the account and organization are the ones you intended, and the repository's settings and permission mode are still loaded.

The Claude Code CLI also supports --model with a full model name. Anthropic identifies the API model as claude-fable-5-1, but the interactive picker is safer for this first verification because it shows what the current account and client actually expose. Use a hard-coded identifier in automation only after your provider and client documentation confirm it for that surface.

Step 4: Run a Read-Only Verification

Do not make a deployment or broad code edit the first test. Give the fresh Fable 5.1 session one bounded request:

Read the repository instructions and summarize the current branch, changed files,
and applicable validation commands. Do not edit files, install dependencies,
change permissions, or contact external services

Verify four observable results:

This test does not prove that Fable 5.1 is better for your workload. It proves that the client, identity, model selection, and repository boundary are coherent enough for a controlled comparison.

Step 5: Compare One Representative Task

Choose a task that is small, reversible, and already has an objective check. Good candidates include explaining one failing test, reviewing a narrow diff, or proposing a plan for a documented bug. Use the same prompt and repository state you used with the previous model.

Compare:

• whether it found the relevant repository instructions;

• time to a useful answer;

• unnecessary tool calls or file reads;

• factual and code-level corrections needed;

• validation commands proposed and actually run; and

• token or cost data returned by your own plan or API surface.

Do not compare models by merging two different changes, changing the prompt between runs, or assuming a higher effort setting is free. Anthropic says Fable 5.1 defaults to High effort in Claude Code. If your Claude Code surface exposes an effort control, test Low or Medium on a non-critical task before adopting a more expensive default. Record the setting with every result.

If Fable 5.1 Is Missing

Work through the boundary in this order:

1. Confirm claude --version resolves the installation you just updated.

2. Restart every Claude Code process and open /model again.

3. Run /status and confirm the expected account and organization.

4. Check whether an organization allowlist or managed setting restricts available models.

5. Check the plan's current model availability and usage policy.

6. If Fable 5.1 still does not appear, use an available model and wait for the provider or administrator to resolve access.

Do not edit enterprise-managed settings, paste API keys into project files, or enable usage credits merely to make the picker change. If Claude Code instead shows Fable 5 requires usage credits, use the separate plan and entitlement troubleshooting guide. That article owns billing-mode and saved [1m] alias failures; this guide owns the Fable 5.1 upgrade and verification path.

Roll Back Without Reverting the Repository

If Fable 5.1 behaves poorly on the representative task, use /model to select the prior approved model for the next session. Re-run /status, preserve the comparison notes, and continue without reverting repository files that were not changed by the model switch.

For unattended scripts, restore the previous model value in the script or its owning configuration and run one dry or read-only test. Do not remove the entire Claude configuration directory. A model choice is independent from the repository's commit history.

If you are evaluating Fable 5.1 against Anthropic's other top coding model, the Claude Opus 5 upgrade guide gives a separate controlled path. Use the model that passes your representative task and cost boundary rather than choosing from launch benchmarks alone.

Verify One Connected Action With Aident Loadout

Once Fable 5.1 passes the repository-only check, test whether its external Action boundary still behaves as expected. Start with the canonical setup instruction:

Follow https://aident.ai/SETUP.md

Then give the session one bounded request:

Check Aident Loadout authentication and Vault status. Discover one connected read-only GitHub issue-search Action, inspect its current schema, and preflight it. If the estimate is free, search for open Claude Code issues mentioning Fable 5.1 and return at most five titles, states, updated times, and canonical URLs. Do not create, edit, label, comment on, close, or react to any issue. Stop before execution if the estimate is not free.

Expected result: one schema-backed, read-only search with no repository edit, provider write, pasted credential, or unapproved cost. This advances the test from local reasoning to one observable integration boundary without turning the model upgrade into an open-ended agent run.

Set up Aident Loadout and verify one read-only Action

Sources

• Introducing Claude Fable 5.1 and Claude Mythos 5.1, Anthropic, September 1, 2026.

• Claude Code CLI reference, Anthropic, reviewed September 1, 2026.

• Claude Code model configuration, Anthropic, reviewed September 1, 2026.

Refresh this guide when Anthropic publishes a minimum Claude Code version for Fable 5.1, changes its model identifier or default effort, changes plan availability, or documents a different upgrade path.