Jan should provide users with an option to configure the reasoning effort level when using reasoning models (like o1, Claude-3.5-Sonnet with reasoning, etc.). This would allow users to control the trade-off between response quality and speed/cost.

Current Behavior

• Users can select different models but have no control over reasoning effort/depth
• Reasoning models use default effort levels without user customization

Proposed Feature

• Add a "Reasoning Effort" slider/dropdown in the model configuration UI
• Support effort levels like "Low", "Medium", "High" or numerical scale (1-5)
• Persist user's effort preference per model
• Apply the effort setting when making API calls to reasoning models

Use Cases

• Users want faster responses for simple queries (low effort)
• Users need thorough analysis for complex problems (high effort)
• Cost-conscious users want to optimize token usage
• Power users need fine-grained control over model behavior

Technical Implementation Areas

• Model configuration interface
• API request parameter handling
• Settings persistence
• Provider-specific reasoning effort mapping

ggml-org/llama.cpp#15266
ggml-org/llama.cpp#15130