Feature: Allow overriding the small/fast model used for subagents

Feature Request

Problem

Currently, Claude Code automatically selects haiku as the model for subagents (Explore, search, and other lightweight tasks). There is no user-facing setting to override this behavior.

While the model field in settings.json controls the main conversation model, users who prefer consistent model quality across all tasks — including subagent operations — have no way to configure which model subagents use.

Proposed Solution

Add a new setting (e.g., smallModelOverride) in settings.json that allows users to override the default small/fast model used for subagents and background tasks.

Example configuration:

{
  "model": "opus[1m]",
  "smallModelOverride": "opus[1m]"
}

Alternatively, this could be an environment variable like CLAUDE_CODE_SMALL_FAST_MODEL.

Use Case

• Users on Max subscription plans who want to maximize output quality across all operations, including subagent tasks
• Scenarios where haiku's capability is insufficient for complex codebase exploration or search tasks
• Users who prefer trading cost/speed for consistency and accuracy

Current Workaround

None. The small model selection is hardcoded internally and cannot be overridden by users.

---

🤖 Generated with Claude Code