Understanding AI Models

A functional overview of AI models on MaaS — purpose of use, basic principle of operation, and the main input data type for each model category.

GreenNode MaaS lets you use leading AI models right away — for chat, coding, image understanding, and document processing — without training or operating your own GPU infrastructure. This page explains what AI models are for, how they work at a basic level, and what kind of input data each model category expects, so you can pick the right model before browsing Available Models.

Purpose of Use

AI models on MaaS are pre-trained deep learning models that automate tasks previously requiring direct human effort: answering questions, summarizing or drafting content, generating and fixing code, reading images, generating images from a description, extracting data from documents, or searching by meaning. Instead of writing hard-coded rules for every case, you describe what you need in natural language (a prompt), and the model reasons out a suitable result.

Principle of Operation

At a basic functional level, an AI model processes a request in three steps:

Input (text / image / audio / PDF)
        │
        ▼
   Tokenize the input data
        │
        ▼
Model predicts/generates output based on
 patterns learned from training data
        │
        ▼
   Output (text / image / vector / JSON...)

Tokenization: Input data (words, image pixels...) is broken down into small units called tokens.
Context-based prediction: A model doesn't "understand" the way a human does — it predicts the most likely next token based on the full context (the prompt, conversation history, system prompt) and patterns learned from a massive amount of training data.
Thinking vs. non-thinking modes: Some models — especially "Reasoning" models such as Qwen, GLM, and DeepSeek — support a thinking mode (step-by-step, chain-of-thought reasoning) that improves accuracy on multi-step problems at the cost of a slower response; non-thinking mode answers faster for simpler tasks.

The more tokens in context (prompt + conversation history), the longer a model takes to process and the higher the cost — see how token-based billing works in Pricing.