Claude Code model configuration changes quickly. As of June 23, 2026, the important durable idea is to use aliases and task shape rather than hardcoding exact model identifiers everywhere.

Useful Aliases

Claude Code supports aliases such as:

• default
• best
• fable
• sonnet
• opus
• haiku
• sonnet[1m]
• opus[1m]
• opusplan

As of the current documentation, opus maps to Opus 4.8 on the Anthropic API, while Claude Platform on AWS, Amazon Bedrock, Google Vertex AI, and Microsoft Foundry can resolve the same alias differently. sonnet maps to Sonnet 4.6 on the Anthropic API and Claude Platform on AWS, and Fable 5 is available for explicit use when the account and retention policy allow it. Check provider-specific model routing before making a team-wide rule.

Defaults

Claude Code defaults vary by account type and provider. That means “use the default model” is a reasonable personal workflow but a weak reproducibility claim. For team processes, specify the alias and date the guidance.

As of 2026-06-23, use `sonnet` for routine implementation and `opusplan` for
planning complex changes before execution.

Opus Plan

opusplan uses Opus for planning and Sonnet for implementation. It is a good fit when the hard part is design, not typing code. Ask for the plan first and review it before allowing broad edits.

Effort and Context

Claude Code also supports effort controls such as /effort and large-context aliases such as sonnet[1m] and opus[1m]. Larger context is not automatically better. It is useful when the task depends on many files, long logs, or generated artifacts. It is wasteful when a targeted test and two source files define the answer.

Fallback Chains

Use fallback models for reliability:

claude --model opus --fallback-model sonnet

Fallbacks are operational policy. They should preserve the intent of the task, not silently downgrade a security review into a cheap rewrite.