About this cheat sheet

A three-page reference to the 11 parameters that control how a large language model generates text, prepared by the Sri Amit Ray Compassionate AI Lab (April 2026 edition). Each parameter has its API name, range, default value, type, a short explanation and practical notes for current models.

What is inside

The five most critical parameters: Temperature, Max Tokens, Top-P (nucleus sampling), Frequency Penalty and Presence Penalty.
Six advanced and specialised parameters: Top-K sampling, Seed, N (number of completions), Stop Sequences, Stream, and Response Format / JSON mode.
Recommended settings by use case: code generation, creative writing, data extraction, chatbot or assistant, summarization, research and analysis, AI agents and tool use, and translation.
Quick reference table: all 11 parameters at a glance with API key, range, default, impact and category.

It applies to GPT-5, Claude 4, Gemini 2.5, Llama 4, Mistral Large 3, Grok-3, DeepSeek-R2 and other major LLM APIs. Parameter names vary slightly between providers, so check the official API documentation for the model you use.

The full explanation of each parameter is in the article LLM Temperature, Top-P, Top-K and the 11 Core Parameters.