Model selection

About

Most users don’t care whether ChatGPT, Claude, or Grok generated their output, as long as that output suits their needs.

That’s why, when manual model selection is available, it’s most often disguised as a parameter like “fast” vs. “deep” responses. This gives users the benefits of different models without making model choice the primary decision. Model details, if shown at all, remain secondary.

Explicit model selection belongs in advanced interfaces for users who need precise control, particularly when AI artifacts become part of their final output (copywriting, image generation). These users understand how model choice affects tone, style, and capabilities.

Since model selection fundamentally shapes AI output, it functions as a persistent parameter.

Types of model selection

1. Automatic selection

While your tool might require various models to provide good value, the system can automatically use different models for different tasks so users aren’t bothered and can focus on their work.

Automatic model selection can, for example, choose:

• Fast, lightweight models for simple actions like Explain or Search

• Advanced models with deeper reasoning for complex tasks like Delegate or Review

2. Manual selection

Creative and technical users need direct model selection to balance:

• Quality and length of the output with utilisation of specific model advantages

• Speed of generation (fast for initial exploration, low for refinement)

• Cost of generation (if included in the plan)

When introducing manual model selection make sure to:

Label models by benefits

Users shouldn't puzzle over "GPT 4.1 versus GPT 4.0" Instead, offer clear benefits like "Fast & Light" versus "Advanced & Thorough." Some products go as far as to skip model names entirely, presenting only capability sliders (speed/accuracy) while handling model selection invisibly.

Default to automatic selection

Power users will find and appreciate manual controls. Everyone else should get intelligent model selection that adapts to their task without their attention

GitHub Copilot groups models by their cost-effectiveness and capability level rather than technical specifications.