Learn one feature at a time
Pick a model
for the task.
Compare Claude model families, understand effort, and avoid confusing capability with permission.
A practical model guide
Start with the model available by default, then change it for a reason. These are starting recommendations, not guarantees. Names and access can vary by plan and provider.
| Family or choice | Good starting use | Example |
|---|---|---|
| Haiku | Small, clear tasks where speed matters. | Summarize a short README or explain a function. |
| Sonnet | Everyday coding and straightforward changes. | Add a small feature with focused tests. |
| Opus | Complex reasoning and difficult debugging. | Trace a subtle bug across several modules. |
| Fable, when available | Large or ambiguous tasks needing sustained investigation. | Investigate a complex failure with explicit scope and review points. |
Names you may see now
As checked October 5, 2026, the direct Anthropic API aliases include Opus 5.5 and Sonnet 5.5; Fable’s mapping can vary by provider. Your picker is the authority for what your account can select. A fixed version name and a moving family alias are different choices.
Opus 5.5 with medium effort is the current default for Pro, Max, Team, Enterprise, and Anthropic API sessions unless a relevant setting overrides it. Claude Platform on AWS, Bedrock, and Google Cloud’s Agent Platform also default to Opus 5.5; Microsoft Foundry defaults to Sonnet 4.5. Organization defaults and saved model or effort settings can differ.
Check and change the selection
In a supported CLI session, /model opens the picker. In graphical interfaces, use the model selector. Review any usage-credit or pricing notice before accepting. Availability does not mean a model is included in your current allowance.
/modelWhat effort changes
Effort changes how much work the model does on a request. This can include investigation and verification, not just waiting longer. Start with the model’s default effort. If it skipped relevant files or checks, more effort may help. If it examined the evidence and still reasoned incorrectly, a more capable model may help.
First check the prompt and context: no model can reliably fix a missing requirement or an inaccessible file. Compare results on one small, representative task before changing your usual setup.
Including ultrathink asks for deeper reasoning on that turn. It adds an instruction without changing the effort value sent to the API.
Usage notices and model switches
Fable may bill usage credits instead of included usage, depending on your plan. Read the picker’s notice before selecting it. Fable models, Opus 5.5, and Opus 5 use safety classifiers. On Fable or Opus 5.5, flagged biology requests can re-run on Opus 5 and flagged cybersecurity requests on Opus 4.8. Opus 5 can fall back to Opus 4.8 for cybersecurity; biology flags end in refusal. Provider configuration and model restrictions can prevent a fallback. Check the notice and current model.
Three separate controls
Model: capability. Effort: amount of work. Permission mode: approval behavior. Increasing one does not automatically improve or change the others. See Permission Modes.
References
- Anthropic. (n.d.-ac). Model configuration. Retrieved September 24, 2026, from https://code.claude.com/docs/en/model-config
- Hallie, L. (2026, July 7). Choosing a Claude model and effort level in Claude Code. https://claude.com/blog/claude-model-and-effort-level-in-claude-code
Was this chapter useful?
Report an errorVotes are shared only if you enable optional measurement in Privacy & Preferences.