[FEATURE] Persistent disable extended thinking setting for Claude Chat (Desktop)

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

Problem: Extended thinking toggle doesn't persist across sessions in Claude Chat (Desktop), even after manually disabling it.

Use case: Users on budget plans (e.g., $30/month) need predictable token usage. Haiku is perfect for routine tasks like checking Jira tickets and data pipeline status, but extended thinking causes unnecessary token burn.

Proposed Solution

Request: Add a persistent setting in Claude Desktop Settings > Model behavior to disable extended thinking by default for specific models (e.g., Haiku), similar to how Claude Code handles MAX_THINKING_TOKENS=0 in settings.json.
Expected: New conversation in Claude Chat defaults to Haiku without extended thinking, persisting across sessions.

Alternative Solutions

Actual: Manual toggle required each time; setting doesn't persist.

Priority

Medium - Would be very helpful

Feature Category

Configuration and settings

Use Case Example

Context: use on a $30/month Claude Pro plan managing daily operational tasks.
Current workflow (manual & inefficient):

1. Open Claude Desktop Chat
2. Start conversation with Haiku model
3. Extended thinking is ON by default
4. Manually toggle extended thinking OFF (every single conversation)
5. Ask Claude to check Jira tickets, verify QTC pipeline refresh status, summarize data anomalies
6. Close conversation
7. Open new conversation → repeat steps 2-4

Problem:

• Extended thinking burns ~3-5x more tokens per task
• Manual toggle each session is tedious
• Token budget depletes faster than necessary
• No persistent way to set "Haiku without thinking" as default

Desired workflow (with persistent setting):

1. Set in Claude Desktop Settings: "disableThinkingForHaiku": true
2. Open Claude Desktop Chat
3. Haiku loads automatically WITHOUT extended thinking
4. Routine tasks (Jira checks, pipeline status, summaries) complete with predictable token usage
5. Can still enable thinking on-demand for complex tasks

Token impact:

Current: ~500 tokens/conversation (thinking overhead)
Desired: ~150 tokens/conversation (routine tasks)
Monthly savings: ~10+ extra conversations on $30 budget

Additional Context

_No response_