Learn AI /Large Language Models /Completion
Completion
Also called: response, generation.
Back to the milestone you came from
In five seconds
What it is
The text a language model produces after a prompt, built token by token from patterns learned during training.
In plain words
A completion is the text the AI writes back after you type something in.
An everyday example
In a writing class in Bonn, two students read the opening lines of a short paragraph: “Heavy rain closed the bridge, so the market traders…” Each student writes a possible ending in their own words. One writes about traders using a different road; the other writes about delays and higher prices. Both endings fit the opening, but neither is guaranteed to be the original paragraph. The teacher gave the same starting text, and the students produced continuations from it. Their endings depend on the words already on the page and on patterns they have learned from reading and writing before.
A student may have intentions, background knowledge and a real sense of the situation. A language model does not work that way. It generates text by calculating likely next tokens from patterns in data and context. Also, a student might remember the exact original paragraph, while a model usually does not retrieve a single source text in that simple way.
Remember thisA completion is the output a language model builds token by token from the prompt and other supplied context.
Understand it
Understand it
Why it matters
If you use chat assistants, you are constantly working with completions, even if you never hear the term. Knowing what a completion is helps you separate your input from the system's output, ask better questions, and review answers more carefully. It also helps in practical decisions: whether to trust the wording, whether to request another version, and whether differences between two outputs came from the same prompt, different settings, or random variation in token choice. In short, the term gives you a clear way to talk about what the model actually produced.
How a completion is produced
After viewing this, a reader should be able to say that a completion is the model's output text, produced step by step from the input context.
01
1. Input arrives
The user provides text, such as a question or instruction.
Draw a rounded rectangle on the left labelled 'Input text'. Inside, show a short example: 'Summarise this note in 2 bullets.' Use a right-pointing arrow to the next box.
02
2. Text becomes tokens
The system converts the text into tokens, which are small text pieces the model works with.
Draw a box labelled 'Tokens'. Show the example split into small pill-shaped units. Add a small note: 'small text pieces'.
03
3. Next token is chosen repeatedly
The model estimates likely next tokens from the context, selects one, appends it, and repeats.
Draw a larger central box labelled 'Repeat'. Inside, show a loop icon and three token pills appearing one after another. Caption: 'predict -> choose -> add'.
04
4. Output text appears
The accumulated tokens form the completion, also called the response or generation.
Draw a box on the right labelled 'Completion'. Show a short output example in two bullet points. Connect from the repeat box with an arrow.
A left-to-right flow diagram with four boxes: input text, tokenised context, repeated next-token selection, and final output. Arrows show how the prompt becomes context, how one token is added at a time, and how those tokens together form the completion.
On hover or tap, each stage highlights and reveals one sentence of explanation; a play button animates token pills moving one by one from the repeat box into the output box.
Describe this without the diagram
The diagram shows a four-stage flow from left to right. First, a person provides input text, for example an instruction asking for a summary. Second, the system converts that text into tokens, meaning small text pieces. Third, the model repeatedly uses the existing context to estimate a likely next token, selects one, adds it to the text, and repeats this cycle. Fourth, the added tokens together form the output text. The key point is that a completion is not the prompt; it is the text produced after the prompt, one token at a time.
With reduced motion: A static four-box flow replaces animation. The third box lists the loop as text: '1 estimate likely next token, 2 choose one, 3 add it, 4 repeat'. The output box shows the final text without moving token pills.
A completion is the output a language model gives after you provide input text. The input might be a question, an instruction, the start of a sentence, or a longer conversation. The output can be one word, one sentence, a paragraph, or many paragraphs.
Inside the model, the input text is split into small pieces called tokens. The model then estimates which token is likely to come next, chooses one, adds it to the text, and repeats. This step-by-step process continues until it reaches a stopping point, such as a length limit or a special end signal.
What comes out is not copied from a hidden answer sheet. It is newly produced text based on patterns from training data and the current context. That is why a completion can be useful, fluent and on-topic, but it can also be wrong, vague or invented.
People often confuse a completion with a prompt. The prompt is what you give the model. The completion is what the model returns. People also use related words such as response or generation. Those are often used loosely, though some systems use them in slightly different ways.
How “Finishing a paragraph in a writing class” maps onto the real thing
| In the story | In the AI system |
|---|---|
| The opening lines on the page | The prompt or prior context given to the model |
| Each student's possible ending | A completion produced by the model |
| What the students learned from earlier reading and writing | Patterns the model learned during training |
| Two different acceptable endings | Different completions can come from the same input |
A worked example
A reviser tests an approved tool on one cleared sentence while working through a translation backlog.
Starting from: Prompt: "Translate into French: The briefing will begin after the afternoon meeting."
- The system tokenises the instruction and source sentence as its context.
- It calculates probabilities for possible next tokens and selects one using its decoding method.
- Each selected token is appended to the context, and the calculation repeats to build the French sentence.
- Generation ends when the model produces a stop token or another configured stopping rule is reached.
Result: Completion: "La réunion d’information commencera après la séance de l’après-midi." The reviser still checks the wording against the source and applicable terminology.
In a United Nations workplace
With a statement for delivery due at close of business and a backlog in the country office translation queue, a reviser uses an approved internal language tool to generate a first-pass translation of two cleared, non-sensitive paragraphs.
The returned draft is the completion. It is generated token by token from the instruction and source passage, rather than retrieved as a finished translation. The reviser then checks terminology, meaning and register before the statement goes for clearance.
Before you try thisSource text containing personal, confidential, security-sensitive or otherwise restricted information must not be entered into a tool that is not approved for that material. Prompts and completions may be retained or exposed. Even with an approved tool, the completion is a draft, not cleared language.
Where you may hear it
- A vendor demo during a procurement meeting
- A translation-service discussion about an approved drafting tool
- An information and communications technology governance board reviewing model settings
- A model evaluation session before an internal tool is released
Avoid confusion
Avoid confusion
Often confused with
Prompt
The text, image or other input that guides a model on what kind of output to produce next.
User prompt
The input a person gives a chat assistant to guide what kind of response it should produce.
Next-token prediction
A language model produces text by repeatedly estimating probabilities for small pieces of text that could come next in the current sequence.
"Understanding" (what models do and do not do)
In language models, apparent comprehension usually means producing text that fits patterns in data and context, not reliably building a grounded grasp of the world like a person does.
Common misconceptions
Common misconception: A completion is the same thing as the prompt.
Correction: They are opposite sides of the exchange. The prompt is the input text you provide. The completion is the output text the model generates from that input. Confusing them makes it harder to diagnose whether a problem came from your instruction or from the model's response.
Common misconception: A completion is always a full, reliable answer.
Correction: A completion is simply the text produced by the model. It may be helpful, partial, off-topic or false. The term describes where the text came from in the interaction, not whether the content is correct.
Common misconception: There is only one correct completion for a prompt.
Correction: Many prompts allow several plausible outputs. Different settings, such as temperature, and different sampling choices can lead to different completions, even when the prompt stays the same.
Quick check
Which statement best describes a completion in a language model system?
A completion is the generated output. The prompt is the input. A model may not check facts, and completions are not simply hidden training passages copied out.
Go deeper
Go deeper
The simple version
A completion is the text the AI writes back after you type something in.
The technical version
In language-model systems, a completion is the generated output sequence conditioned on the supplied context. The input text is tokenised, the model computes a probability distribution over possible next tokens, one token is selected by a decoding method such as greedy choice or sampling, and that token is appended to the context. The process repeats until a stop token, stop rule or length limit is reached. In everyday product language, completion, response and generation are often used interchangeably, though interfaces may define them differently. The term should be distinguished from the prompt, which is the conditioning input, and from next-token prediction, which is the mechanism used to produce the completion.
Reviewed external resources
EncyclopaediaWikipedia search: Completion (opens in a new tab)Wikipedia
This Wikipedia search results page helps readers see that 'completion' has several meanings, so it suits someone who first wants to disambiguate the term before focusing on AI.
search results, not one pagelink checked 2026-07-31
CourseLLM Course (opens in a new tab)Hugging Face
This course suits readers who want a structured introduction to how language models generate outputs, including the mechanics behind completions.
Deeperlink checked 2026-07-31
VideoLarge language models, explained visually (opens in a new tab)3Blue1Brown
This video suits visual learners who want an intuitive explanation of how large language models produce text step by step.
Intermediatelink checked 2026-07-31
VideoVideo explanations: Completion (opens in a new tab)YouTube search
This YouTube search results page suits readers who want to compare multiple video explanations of completion at different difficulty levels.
search results, not one pagelink checked 2026-07-31
DiscussionCommunity discussion: Completion (opens in a new tab)Reddit search
This Reddit search results page suits readers who want to see common questions, confusions and informal discussions about completion, while remembering that community answers vary in quality.
search results, not one pagelink checked 2026-07-31
Where this came from
Drafted with a language model through an offline pipeline, checked against this site's schema, and every external link above fetched to see whether it answers. Treat it as a starting point and check anything that matters against the primary sources. How this site is built and what that means.
The one thing to remember
A completion is the output a language model builds token by token from the prompt and other supplied context.