To use Claude Fable 5.1 in Claude Code, update to Claude Code 2.1.255 or later, then run /model fable inside a session or launch a new one with claude --model fable. Fable 5.1 supports five effort levels—low, medium, high, xhigh, and max—and its model default is high. Use /effort to change the balance between speed, token use, and reasoning depth.
After selecting the model, check /status to confirm the account and model route actually in use. If Fable is missing or a request fails, investigate the Claude Code version, current credential, organization restrictions, provider configuration, and Usage Credits in that order. Do not assume that a paid plan, a visible picker row, or a working API key proves every other access condition.
Facts verified September 9, 2026. Model aliases, minimum versions, effort controls, availability, and billing rules can change. Treat Anthropic's current model configuration documentation, the live
/modelpicker,/status, and your account or organization controls as the final authority.
The 60-second setup
First check and, if necessary, update Claude Code:
Then start a new Fable 5.1 session at the default high effort:
Or switch an existing interactive session:
Anthropic's current Claude Code model configuration guide says the fable alias resolves to Fable 5.1 unless an administrator or provider configuration overrides that alias. It also documents 2.1.255 as the minimum Claude Code version for Fable 5.1.
This guide is about the Claude Code CLI experience. For API rates, prompt caching, and the model's broader coding-agent fit, use the separate Claude Fable 5.1 model guide.
Select Fable 5.1 explicitly
Fable is not the account-type default on any Claude plan or provider, so choosing it should be an explicit step.
Switch the current session
Run:
This asks Claude Code to select the current model behind the fable alias. Running /model without an argument opens the picker instead. If the conversation already has output, Claude Code may ask you to confirm the switch because the next response must read the existing history under the newly selected model.
Start a clean Fable session
Run:
This is the clearer route when the task should begin with a clean context and a known model family. Add --effort high, or another supported level, when you want the effort choice to be explicit for that session.
The alias is convenient but intentionally moves with Anthropic's recommended Fable release. At the time of verification, it resolves to Fable 5.1. If an audited workflow must pin an exact version, verify the current full model ID and provider-specific mapping before storing it in settings. An Anthropic API ID, an Amazon Bedrock identifier, and a gateway alias are not automatically interchangeable.
Choose among the five effort levels
Fable 5.1 supports all five current effort levels. high is the model default when no saved setting, command-line flag, environment variable, or organization policy overrides it.
To choose interactively, run /effort and use the slider, or set a level directly:
You can also adjust the effort slider while selecting a supported model in /model, or set one session at launch:
Use /effort auto to clear the saved effort for the active model and return to its default behavior. Claude Code normally saves low, medium, high, and xhigh per model for later local sessions. max is session-only unless it is imposed through the relevant environment variable. An organization can also cap which levels members may use.
Anthropic's current effort guidance recommends starting Fable 5.1 at high, then stepping up only for capability-sensitive work or down after evaluations show that quality holds. Effort is a behavioral control, not a guaranteed token budget, and the same label is calibrated separately for each model.
Adaptive thinking cannot be turned off on Fable 5.1
Fable 5.1 always uses adaptive thinking. That means the model decides whether and how deeply to reason on each step, guided primarily by the selected effort level. It does not mean every simple turn must produce a long visible reasoning block.
Claude Code's usual thinking toggle, alwaysThinkingEnabled, and MAX_THINKING_TOKENS=0 do not disable thinking on Fable 5.1. Lower /effort when a routine task does not justify the default depth; raise it only when the task and acceptance checks justify the extra work.
Keep two similarly named controls separate:
- Claude Code
/effortand--effortconfigure the interactive or launched CLI session described in this guide. - Claude API per-message effort is an API feature for changing
output_configon individual messages. It is not a slash-command recipe and should not be pasted into Claude Code as if it were one.
This distinction matters when troubleshooting. A code example written for direct API calls will not prove what the Claude Code interface saved for the current session.
If Fable 5.1 is missing or does not run
Work through these checks in order. Record the exact error before changing several settings at once.
1. Check the Claude Code version
Run claude --version. Fable 5.1 requires 2.1.255 or later. If the client is older, run claude update, open a new session, and try /model fable again.
If more than one Claude Code installation exists, check the version from the same terminal, IDE, Desktop surface, or remote environment that will run the task. Updating one binary does not update another copy elsewhere.
2. Try the typed alias, not only the picker
On the Anthropic API, the picker lists Fable only after the server reports it available to the organization. Anthropic documents that a typed /model fable selection checks availability with the server directly, so it can succeed even when the picker row is absent.
If the typed command is rejected, keep the message. It helps distinguish a missing entitlement, an organization restriction, an unsupported client, and an unrecognized provider mapping.
3. Confirm the account and credential
Run /status. Check which Claude account, organization, and authentication route are active. A subscription identity, a Claude Console API key, a cloud-provider credential, and a gateway are different billing and policy boundaries.
An environment variable or managed configuration can take precedence over the account you expected. Do not publish API keys, tokens, or unredacted credential files while asking for help.
4. Check organization restrictions and provider mappings
An Enterprise administrator can restrict model access by organization or custom role. Managed availableModels settings can also hide or reject a model. If every Fable version is blocked, the family alias cannot select one.
Third-party providers and LLM gateways may use their own model identifiers or override ANTHROPIC_DEFAULT_FABLE_MODEL. Ask the administrator which exact Fable deployment is enabled and whether the alias points to it. Do not copy an Anthropic API ID into a provider configuration without checking that provider's mapping.
5. Check Usage Credits and the current plan route
Depending on the plan and seat tier, a Fable request may bill to Usage Credits rather than included plan limits. Claude Code can label the picker row “Requires usage credits” and, in an interactive session, ask for consent before charging them. Organization-billed Enterprise members can follow different prompt behavior.
Being allowed to select the model does not guarantee that the active account has a usable balance or approved spend. Check the current account controls and the owner of the bill. The Claude Code pricing guide separates subscription allowance, Usage Credits, and Console API billing. If the failure is a limit or balance message rather than model selection, follow the Claude Code usage-limit recovery guide.
If the problem remains, collect the client version, command used, provider, account and organization names, non-secret /status output, and full error text. That evidence is more useful than repeatedly reinstalling the CLI or switching credentials.
Match effort to a real task
Anthropic positions Fable 5.1 for hard, long-running work. That is a vendor description, not independent proof that it is the best or cheapest route for your repository.
Start with one bounded task whose result can be reviewed—for example, a cross-module defect investigation, a difficult architecture review, or a migration with explicit tests. Keep the starting code, instructions, tools, permissions, and acceptance checks stable while comparing effort levels.
Measure accepted outcomes rather than answer length:
- checks passed and defects missed;
- repair turns and reviewer time;
- elapsed time and interruptions;
- input, output, and cached-token usage where visible; and
- whether the final change stayed inside the requested scope.
Use high as the baseline. Try medium for routine work and compare quality; try xhigh or max only when a difficult task leaves meaningful room for improvement. A one-line edit does not become better merely because the session uses the most expensive reasoning setting.
Where Agent.Space fits
Claude Code is an Agent harness. Fable 5.1 is a model. Agent.Space provides a separate Agent and persistent Workspace product layer. These parts can support the same project, but they are not interchangeable subscriptions or guarantees of model access.
Agent.Space is independent from Anthropic. This article does not claim that Fable 5.1 is currently available with Claude Code on Agent.Space. The live Agent.Space selector is the source of truth for supported Agent and model combinations; the Claude Code on Agent.Space guide explains the surrounding Workspace workflow and product boundaries.
If that persistent Workspace model fits your project, start with Agent.Space, confirm the current Agent, model, and price shown for your account, and give the first session one small task with an observable acceptance check.
FAQ
What command selects Fable 5.1 in Claude Code?
Run /model fable inside a session, or start a new one with claude --model fable. As verified on September 9, 2026, the fable alias resolves to Fable 5.1 unless an administrator or provider configuration overrides it.
What Claude Code version does Fable 5.1 require?
Claude Code 2.1.255 or later. Run claude --version, then claude update if necessary.
What is the default effort for Fable 5.1?
The model default is high. A saved per-model preference, --effort, /effort, an environment variable, or an organization policy can change the effective level.
Can I turn thinking off on Fable 5.1?
No. Fable 5.1 always uses adaptive thinking. Use an appropriate effort level to steer depth; Claude Code's usual thinking-off controls have no effect on this model.
Why is Fable missing from /model?
The server may not report it as available to the active organization, the client may be older than 2.1.255, an administrator may restrict it, or a provider mapping may be missing. Try typed /model fable, then check the version, /status, organization policy, provider, and Usage Credits.
Is Claude Code /effort the same as API per-message effort?
No. They influence the same general capability-cost tradeoff but belong to different interfaces. This guide uses Claude Code's /effort and --effort; direct API applications configure per-message effort in their request payloads.