Common Questions About Model Selection

Does Dessix Support Custom Third-party Models?

Answer: No

Design Philosophy

To reduce the cognitive burden on users during usage, Dessix has built an intelligent framework for understanding tasks and intentions.

Intelligent Model Selection

• Automatic Recognition: Dessix automatically identifies task types and user intentions
• Optimal Matching: The most suitable model for the current task and intention helps you complete tasks
• One-shot Results: Provides results as close to your goals as possible in one attempt

How to Choose Models?

Intelligent Selection Mechanism

In Dessix, model selection is intelligent and implicit:

• Dessix automatically selects the best model based on task type for responses
• No manual user selection needed, reducing decision burden
• Intelligent matching based on task characteristics and context

Explicit Model Specification

We also support users specifying models in specific scenarios:

Specifying in Conversations

When conversing or triggering Actions, you can explicitly declare which model you want to use:

Example 1: Direct Model Specification

Use claude-4-sonnet to complete this task, ......

Example 2: Task Type Hints

This is a programming task, ......

The system will automatically switch to a model suitable for programming (such as claude-4-sonnet).

Code Assistant Action

This approach is particularly useful in code assistants, where you can choose appropriate models based on task complexity.

Built-in Model List

Dessix currently includes the following high-quality models:

Claude Series

• claude-3.5-sonnet - Balanced performance and efficiency
• claude-3.7-sonnet - Enhanced version
• claude-4-sonnet - Latest flagship model

DeepSeek Series

• deepseek-r1 - Reasoning-optimized version
• deepseek-chat - Conversation-optimized version

Gemini Series

• gemini-2.0-flash - Fast response version
• gemini-2.5-pro - Professional version

GPT Series

• gpt-4o - OpenAI flagship model
• gpt-4.1 - Latest version
• gpt-4.1-mini - Lightweight version
• gpt-4.1-nano - Ultra-lightweight version

Other Models

• grok-3 - X.AI flagship model
• grok-3-mini - Lightweight version
• o3-mini - Reasoning-optimized model
• o3-mini-high - High-performance reasoning model
• qwen-max - Alibaba Cloud flagship model

Design Considerations

Why Not Expose Model Selection?

Avoid Becoming a Playground Tool

• If model selection is exposed, Dessix might become a complex Playground tool
• Would need to explain to users which models are suitable for what tasks
• Model SOTA is constantly changing, which would increase users' cognitive burden

Reduce Cognitive Load

• Users don't need to understand the characteristics and applicable scenarios of various models
• Focus on the task itself rather than tool selection
• Simplify usage flow and improve efficiency

Transparency Needs for Advanced Users

For advanced users who need to know which models are being used:

Recommended Approach

• Clearly express intentions and requirements when asking questions
• Can specify through model codes (from the above list)
• System will match based on specifications

Usage Tips

• Describe task types: programming, writing, analysis, etc.
• Specify quality requirements: fast response vs high-quality output
• Clearly state model preferences: such as "use claude-4-sonnet"

Best Practices

General Users

• Directly describe your needs and tasks
• Let Dessix intelligently choose the most suitable model
• Focus on result quality rather than model selection

Advanced Users

• Include model preferences in task descriptions
• Hint at expected model capabilities through task types
• Use explicit specification features for precise control

Through this design, Dessix maintains ease of use while providing sufficient control for advanced users.