Deep Reasoning (Thinking Mode)

Google Gemini 3 models feature built-in reasoning mechanisms (Chain of Thought). AMC WebUI provides granular controls and real-time visualization for this process.

Control Dimensions

Under Settings -> Generation Settings:

1. Thinking Level

Select between 4 reasoning depths:

Minimal: Basic internal consistency check (gemini-3.5-flash-lite and gemini-3.6-flash; 3.7 and 3.8 Flash do not support Minimal per API constraints).
Low: Light reasoning, suitable for straightforward queries.
Medium: Balanced exploration for general problem solving.
High: Full-power deep reasoning for complex architecture, logic, and mathematics.

💡 Quick Toggle: Type /fast in the chat input to toggle between Low and High thinking levels.

2. Thinking Token Budget

Adjust the slider from 128 to 32,768 tokens.
Set to -1 for automatic allocation determined by the model based on question difficulty.

Chain of Thought Visualization

In conversation view:

Streaming Steps: Watch the model examine hypotheses and self-correct before producing final answers.
Metrics: View elapsed reasoning time and token consumption upon completion.
Thought Signatures: Preserves internal thought tokens across turns to sustain context continuity.

Automated Thought Translation

Gemini models frequently formulate internal reasoning in English. AMC WebUI includes an automatic translation engine:

Turn on “Translate Thought Process” in settings.
Lightweight gemini-3.5-flash-lite translates English thoughts into your chosen UI language in the background without delaying the primary streaming response.