Thinking Modes

Thinking modes control how much reasoning effort Atom applies when processing your request. Higher modes use more deliberate reasoning for complex tasks, while lower modes prioritize speed.

Available Modes

Auto (Recommended)

Atom automatically assesses the difficulty of your request and chooses the appropriate reasoning level. This is the default and works well for most cases.

When it’s useful: Almost always. The auto-assessment is accurate enough that you rarely need to override it.

Low

Minimal reasoning, optimized for speed and cost. Best for straightforward, repetitive tasks.

When it’s useful:

  • Simple batch operations (rename layers, reset properties)
  • Quick property adjustments
  • Tasks you’ve done before in the same session

Tradeoff: Faster and cheaper, but may miss nuance on complex requests.

Mid

Balanced reasoning for moderately complex tasks.

When it’s useful:

  • Multi-step workflows with dependencies
  • Rigging and control setups
  • Expression creation with specific logic

Tradeoff: Slightly slower than Low, but handles complexity better.

High

Maximum reasoning for challenging or ambiguous requests.

When it’s useful:

  • Complex animation sequences with timing constraints
  • Intricate expression logic across multiple layers
  • Tasks requiring careful analysis of layer relationships

Tradeoff: Slower and more expensive, but solves harder problems more reliably.

Changing the Mode

Set your thinking mode in Settings → General → Thinking Mode, or use the dropdown in the chat input area.

Cost and Speed

Higher thinking modes use more compute, which means:

  • Longer response times – Complex reasoning takes more time
  • Higher token usage – More reasoning consumes more API credits

For most work, leaving it on Auto gives you the best balance. Atom will use higher reasoning only when the task demands it.

When to Override Auto

Consider manually selecting a mode when:

  • You know the task is simple and want guaranteed speed (Low)
  • You’re debugging a stubborn issue and want maximum reasoning (High)
  • You’re iterating rapidly and want consistent speed (Low or Mid)