[FEATURE] Feature Request: Model-conditional configuration settings (e.g., thinking mode defaults per model)

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Description

I'd like the ability to configure different default settings based on which model is active. The primary use case is setting different thinking mode defaults:
Opus: Default to thinking off (extended thinking adds latency without proportional benefit for this already-capable model)
Sonnet: Default to thinking on (benefits significantly from extended thinking)

Proposed Solution

Proposed Solution

One or more of these approaches:

Option A: Conditional settings syntax in settings.json

{
"env": {
"MAX_THINKING_TOKENS": {
"default": "10000",
"claude-opus-4-5-20251101": "0",
"claude-sonnet-4-5-20250929": "16000"
}
}
}

Option B: Model-specific configuration blocks

{
"modelDefaults": {
"opus": {
"MAX_THINKING_TOKENS": "0"
},
"sonnet": {
"MAX_THINKING_TOKENS": "16000"
}
}
}

Option C: Expose model to hooks

Add CLAUDE_CURRENT_MODEL to hook environment variables, allowing users to implement custom logic:

#!/bin/bash
#.claude/hooks/model-thinking.sh
if [[ "$CLAUDE_CURRENT_MODEL" == "opus" ]]; then
export MAX_THINKING_TOKENS=0
else
export MAX_THINKING_TOKENS=16000
fi

Alternative Solutions

Current Workaround

None that's automatic. Manual options:
Press Tab after every /model switch
Use separate terminal sessions with different MAX_THINKING_TOKENS env vars
Include/omit "think" in every prompt

Priority

Medium - Would be very helpful

Feature Category

CLI commands and flags

Use Case Example

Use Case

Different models have different cost/latency/capability tradeoffs. Extended thinking significantly improves Sonnet's output quality but adds latency and cost to Opus where the base model is already highly capable. Currently, users must manually toggle thinking with Tab every time they switch models, which is easy to forget.

Additional Context

Additional Context

This pattern could extend beyond thinking mode to other model-specific defaults (temperature, system prompt additions, etc.) as the multi-model workflow becomes more common.