Models / Reference

Models

Match the model to the moment.

Finkkle models are tuned for different balances of speed, depth, and tool use. Start with the smallest model that can reliably finish the job.

Choosing a model

Use fast models for classification, summaries, and short transformations. Use reasoning models for ambiguous requirements, multi-step plans, and technical decisions. For large inputs, prioritize context capacity and clear source boundaries.

Prompting for quality

  • State the outcome, audience, and format first.
  • Include only the context the model needs to decide.
  • Ask for assumptions and uncertainties when the answer matters.
  • Break complex work into reviewable steps.
TipModel choice is only half the result. A crisp objective and explicit constraints usually improve quality more than switching models.

Limits and safety

Models can be wrong, incomplete, or overly confident. Verify important facts, inspect generated code, and keep secrets out of prompts and attachments.