What Is Word Completion?
Word completion is a stateless API call. You send a single prompt, and the model returns a continuation without retaining any memory of previous interactions. Key characteristics:- Single-shot responses
- No conversation history
- Simpler, lower token usage

What Is Chat Completion?
Chat completion is stateful. You maintain a record of all messages in amessages array, allowing the model to build on past user and assistant exchanges.
How Chat Completion Works
- You create an ordered
messageslist. - Each entry has a
role: system, user, or assistant. - You submit the full history each time you call the API.
- The model returns a response that accounts for all previous context.
Roles in the messages Array
| Role | Description | Example |
|---|---|---|
| system | Sets global instructions or persona | “You are a physics professor.” |
| user | Represents the human’s input at each turn | “Explain quantum mechanics in simple terms.” |
| assistant | Model-generated responses that continue the conversation context | “Quantum mechanics is the study of matter at very small scales…” |

- Use word completion for simple text continuations, code generation, or single-turn tasks.
- Use chat completion for multi-turn conversations, contextual assistants, and applications that require stateful interaction.
Sample Chat Completion Request
Comparison at a Glance
| Feature | Word Completion | Chat Completion |
|---|---|---|
| Context Handling | Stateless | Stateful via messages array |
| Best for | Single-turn prompts | Multi-turn conversations |
| Token Efficiency | Fewer overhead tokens | More overhead tokens for history |
| Recommended Model | GPT-3.5-Turbo | GPT-3.5-Turbo |