Feature Request: Per-Prompt Model Override Syntax (//s, //o, //h)

Feature Request: Per-Prompt Model Override Syntax

Problem

Currently, switching models in Claude Code with /model changes the entire session's model. For cost-conscious workflows where Sonnet is sufficient 80% of the time but Opus is needed occasionally, users must:

1. /model → switch to Opus
2. Ask the complex question
3. /model → switch back to Sonnet

This is tedious and disrupts flow.

Proposed Solution

Add a lightweight per-prompt model override syntax that uses a model for a single message only, without changing the session default:

//s quick analysis task    → uses Sonnet for this prompt only
//o complex refactoring     → uses Opus for this prompt only  
//h fast lookup             → uses Haiku for this prompt only

After the response, the session reverts to whatever model was set with /model.

Benefits

• Cost optimization: Default to Sonnet, use Opus only when needed
• Workflow efficiency: No manual model switching back and forth
• Cleaner sessions: Model changes are explicit per-prompt, not global side effects

Related Issues

• #4016 (closed as duplicate): Proposed [model:opus] syntax
• #4807 (closed): Broader granular model control proposal

This is a simpler, more focused version of those requests — just a quick way to override the model for one prompt without affecting the session.

Use Case

User running a long coding session on Sonnet (default) occasionally hits a complex architectural decision that needs Opus-level reasoning. Instead of switching the entire session to Opus and back, they can ask:

//o design the error handling strategy for this async state machine

Next prompt automatically reverts to Sonnet.