Thinking Modes
Enable deep reasoning for complex tasks. Synoppy supports both extended thinking and adaptive thinking depending on the model.
How Thinking Works
Thinking allows the AI to "think out loud" before responding. Instead of jumping straight to an answer, the model explores the problem space, considers alternatives, and reasons through complex decisions.
Extended Thinking
Used by Sonnet 4.5, Opus 4.5, and Haiku 4.5. You set a thinking budget (max tokens for reasoning) and the model uses up to that amount. Good when you want to control how much reasoning the model does.
Adaptive Thinking (Opus 4.6 & Sonnet 4.6)
Used by Opus 4.6 and Sonnet 4.6. The model automatically decides how much to think based on the complexity of your request. No budget configuration needed - it adapts on its own.
Enabling Extended Thinking
Toggle during a session:
/thinkingUse keyboard shortcut:
Enable by default in config:
synoppy config set thinkingEnabled trueSet thinking budget (max tokens for thinking):
synoppy config set thinkingBudget 16000When to Use Extended Thinking
Complex debugging
Multi-step debugging that requires understanding system architecture
Architecture decisions
Designing systems with multiple interacting components
Algorithm optimization
Analyzing and improving complex algorithms
Code review
In-depth analysis of code for issues and improvements
Thinking Budget
For extended thinking models (Sonnet, Opus 4.5, Haiku), the thinking budget controls how many tokens the model can use for reasoning. Higher budgets allow for deeper analysis but take longer and cost more.
Note: Opus 4.6 and Sonnet 4.6 use adaptive thinking and ignore the budget setting - they automatically determine how much to think based on task complexity. Gemini models have their own built-in thinking that is always on.
| Budget | Best For |
|---|---|
5,000 | Quick analysis, simple decisions |
10,000 | Standard tasks (default) |
16,000 | Complex debugging, architecture |
32,000 | Deep analysis, research tasks |
Model Support
Thinking is available on select models:
Tips
Don't overuse it: Extended thinking adds latency. For simple tasks like "add a console.log", it's overkill.
Pair with context: Provide relevant files and context so the model has good material to reason about.
Use Opus 4.6 for auto-thinking: If you don't want to manage thinking budgets, Opus 4.6's adaptive thinking automatically scales reasoning to match the task. No configuration needed.
Start simple: Try Haiku 4.5 with thinking enabled first — it's fast and affordable. Scale up to Sonnet or Opus only if the task needs deeper reasoning.