What does LLM mean?
LLM stands for large language model: a language model with many parameters, trained on large amounts of text data, that generates text from the input it receives.
Researchers coined the term for pre-trained language models of significant size (Zhao et al.). Common uses include answering questions, summarizing information and generating human-like text (Anthropic glossary).
Large compared with what?
“Large” is relative to earlier, smaller language models: more parameters and the ability to use more context (Google Machine Learning Crash Course). The Zhao et al. survey notes that once models pass a certain scale, they not only perform better but also show abilities that smaller models lack.
There is no agreed parameter count that makes a model an LLM. As a dated example, GPT-3, published in 2020, had 175 billion parameters (Brown et al.). That was its size at the time; it does not mean current models are in the same range.
LLMs and tokens
An LLM reads and writes text in tokens. Text is split into a sequence of tokens before the model processes it (Anthropic glossary). A token can be a word, part of a word or a single character (Google Machine Learning Crash Course).
To generate an answer, an LLM predicts the next token or sequence of tokens, appends the prediction to the preceding text and keeps going. How many tokens the same passage becomes therefore depends on the model and the language. There is no shared official figure for how many Chinese characters one token covers; it varies by model and language.
A prediction is not a verified fact
An LLM learns patterns and relationships in its data, so no output should be treated as a checked fact. A plausible passage can still contain an AI hallucination.
The model, chat products and search are different layers
An LLM is the model itself. A chat interface can connect it to search, calculation or file tools, so whether an answer drew on the web depends on which features the product had enabled and whether a tool was used for that request (Microsoft Copilot support). Knowing that an LLM was used does not tell you.
AI search combines information retrieval with generated answers.
Common questions
What does LLM mean?
LLM stands for large language model: a language model with many parameters, trained on large amounts of text data, that generates text from the input it receives.
How many parameters does it take to be an LLM?
There is no agreed threshold. The Zhao et al. survey gives tens or hundreds of billions of parameters as examples and limits its own scope to models above 10 billion, but notes that there is no formal consensus on the minimum.
How many Chinese characters is one token?
There is no fixed ratio. How many tokens a passage becomes varies by model and language; see the Token entry.
Does a chat product built on an LLM always search the web?
No. It depends on whether the product used a tool such as search for that request; searching is not something the LLM itself does.
Sources and scope
- Zhao et al.: A Survey of Large Language Models (arXiv 2303.18223)
- Google for Developers: LLMs: What's a large language model?
- Anthropic: Claude Platform Glossary
- Brown et al.: Language Models are Few-Shot Learners (arXiv 2005.14165)
- IBM: What are large language models?
- Microsoft Support: information used to answer a Copilot prompt
Sources as checked for the source-language entry on 2026-10-27; this English edition adds no new claims.