[DOCS] Conflicting information regarding Plan subagent model vs. Opus 4.5 reasoning
Documentation Type
Missing documentation (feature not documented)
Documentation Location
https://code.claude.com/docs/en/sub-agents#plan-subagent https://code.claude.com/docs/en/model-config#opusplan-model-setting
Section/Topic
Built-in subagents > Plan subagent and Model Aliases (opusplan)
Current Documentation
The "Subagents" page under "Built-in subagents" states:
Plan subagent Model: Uses Sonnet for more capable analysis
The "Model configuration" page states:
opusplanmodel setting In plan mode - Usesopusfor complex reasoning and architecture decisions In execution mode - Automatically switches tosonnetfor code generation and implementation
What's Wrong or Missing?
There is a direct contradiction and a lack of clarity regarding which model handles the "Planning" phase in versions 2.1.0 and higher:
- Conflicting Models: The built-in subagents documentation explicitly says the Plan subagent uses Sonnet. However, the
opusplanalias and the "Models overview" describe Opus 4.5 as the smartest model for premium reasoning. - Override Ambiguity: As of version 2.1.0, the Plan subagent is triggered automatically by the
/plancommand orEnterPlanModetool. If the built-in agent is hardcoded to use Sonnet (as stated in the Subagents doc), it is not using the "smartest" reasoning available (Opus 4.5). - Redundancy/Logic Gap: If the Plan subagent is intended to be a "Software Architect," it is confusing why it would default to a less intelligent model than Opus 4.5, especially when the documentation for
opusplansuggests that Opus is the preferred choice for the planning phase. - No Override Path: There is no documented way to tell the built-in Plan subagent to use Opus reasoning specifically, since subagents operate in isolated contexts and the
/planshortcut doesn't currently accept model overrides.
Suggested Improvement
The documentation should be updated to clarify the relationship between the built-in Plan subagent and the opusplan model alias:
- Clarify Default: State clearly whether the Plan subagent uses Sonnet by default or if it respects the
opusplanalias/Opus 4.5 when configured. - Update Subagent Table: If the Plan subagent can use Opus, update the "Built-in subagents" table to reflect that it uses the highest available reasoning model (Opus 4.5) when in
opusplanmode. - Instruction for Best Reasoning: Provide a specific command or configuration (e.g.,
ANTHROPIC_MODEL=opusplanorclaude --model opusplan) that ensures the Plan subagent utilizes Opus 4.5 for the architectural research phase. - Address Hardcoding: If the built-in subagent is indeed locked to Sonnet, provide a justification for why the "smartest" model isn't used for the planning phase, or provide a way to override it.
Impact
High - Prevents users from using a feature
Additional Context
Users frequently use /plan for complex refactors (touching 10+ files). If this phase is handled by the Sonnet-based Plan subagent instead of Opus 4.5, the architectural integrity of the plan may be lower than expected based on the "Models overview" performance claims.
This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗