Thinking sets how long the AI reasons before it answers. More thinking can give better answers on hard tasks, but it uses more credits.
Where to change it
Click the model name under the text box.
Click Thinking.
Pick a level.
If you don't see Thinking, the model you picked doesn't have it. Gemini 3.8 Flash, DeepSeek V4.1 Flash, DeepSeek V4 Pro, Grok 4.7 and Grok 4.5 have no thinking levels.
Which levels each model has
Claude 5 Sonnet and GPT-5: Instant, Low, Medium, High
Claude Opus models, GPT-5.6 Terra, GPT-5.5 and GPT-5.4 Mini: Instant, Low, Medium, High, Extra High
Claude Opus 5.5 always thinks a little, even on Instant.
Instant
Best for: quick, everyday tasks.
The AI answers right away with no extra reasoning. Use it for simple questions, short scripts and quick edits.
Example: "Write me a caption for this image"
Avoid when: the task is complex.
Low
Best for: slightly more involved tasks.
A little thinking before it answers. Use it for writing and basic content.
Example: "Write me a product description for my new feature"
Avoid when: the task is technical or has many parts.
Medium
Best for: moderately complex tasks.
More reasoning. Use it for longer content or when accuracy matters. Poppy warns that Medium can use 2-3x more credits.
Example: "Help me write a full email sequence for my launch"
Avoid when: the task is simple.
High
Best for: hard problems.
The AI works through the problem in depth. Try it when other levels didn't work. Poppy warns that High can use 2-3x more credits.
Example: "Help me plan the entire structure of my course"
Avoid when: the task is simple.
Extra High
Best for: the hardest tasks.
Maximum reasoning. Only on the Claude Opus models, GPT-5.6 Terra, GPT-5.5 and GPT-5.4 Mini. It uses the most credits.
Example: "Help me design a full business strategy"
Avoid when: almost anything else.
Rule of thumb
Start with Instant. If the result isn't good enough, go up one level at a time. To save credits, see Troubleshooting High Credit Usage.