What Is a Large Language Model (LLM)?
A large language model (LLM) is a machine learning model, usually a neural network, trained on a vast amount of text for language tasks, especially generating language. It builds each answer by repeatedly predicting the next word in a sequence (Large language model, Wikipedia) (What are large language models (LLMs)?, IBM Think).
Last updated: November 2026
How a large language model works
Text goes into and comes out of the model as tokens. A token can be a whole word, a subword or a single character, and most modern language models split text by subwords: chunks of text that carry meaning, with prefixes and suffixes sometimes kept as separate pieces (Introduction to Large Language Models, Google for Developers). By modeling the statistical patterns of tokens, a language model builds powerful internal representations of language (Introduction to Large Language Models, Google for Developers).
LLMs are typically built on the transformer neural network architecture, and generative pre-trained transformers (GPTs) are one type of LLM (Large language model, Wikipedia). Size is counted in parameters: the neurons in the network's layers and the weights and biases between them (Large language model, Wikipedia). GPT-3, for example, has 175 billion parameters (Language Models are Few-Shot Learners, Brown et al., arXiv).
No fixed parameter count makes a model large; the label is inherently vague (Large language model, Wikipedia). Models keep growing in size and expense: GPT-2, released in 2019 with 1.5 billion parameters, cost an estimated $50,000 to train, and PaLM, launched three years later with 540 billion parameters, cost an estimated $8 million (2023 State of AI in 14 Charts, Stanford HAI).
Why it matters for hotel distribution
Travelers already plan trips with these models. Nearly four in 10 US travelers used generative AI to research a trip in the 12 months before November 2025, through platforms like ChatGPT or AI search features such as Google's AI Mode, an increase of 11 points in one year (Search slips, AI surges, Phocuswright). The vast majority of them made at least one travel decision from the answers they received, and AI has reached parity with travel publications and supplier content as a resource (Search slips, AI surges, Phocuswright). Trust is still building: friends and family remain the most influential resource, and traveler reviews are trusted more than most other sources (Search slips, AI surges, Phocuswright).
Booking.com put an LLM into its booking path in June 2023 with its AI Trip Planner, partly powered by OpenAI's ChatGPT API and built into the accommodation booking experience in the Booking.com app, starting with a select group of US Genius loyalty members (Booking.com launches ChatGPT trip planning feature, Marketing Dive).
On October 6, 2025, OpenAI launched apps inside ChatGPT, and Booking.com and Expedia were among its seven pilot partners, available in markets where their services are offered, starting in English, and not in the European Economic Area, Switzerland or the UK at launch (Introducing apps in ChatGPT and the new Apps SDK, OpenAI). OpenAI's launch example shows a request for a hotel in Paris for two adults with parking, answered by Booking.com with hotel results, photos, prices and amenities inside the chat (Introducing apps in ChatGPT and the new Apps SDK, OpenAI). Apps built this way can reach over 800 million ChatGPT users, by OpenAI's count (Introducing apps in ChatGPT and the new Apps SDK, OpenAI). The apps are built on the Model Context Protocol (MCP), the open standard that lets ChatGPT connect to outside tools and data (Introducing apps in ChatGPT and the new Apps SDK, OpenAI).
How a hotel or rental gets named in those answers is a separate subject, covered in our guide to AI search optimization for hotels and in the glossary entries on generative engine optimization and answer engine optimization.
Large language model compared with generative AI, chatbots and MCP
| Large language model | Generative AI | AI chatbot or concierge | Model Context Protocol (MCP) | |
|---|---|---|---|---|
| What it is | The underlying model | The broader category of AI that generates content | The product a guest talks to | A connection standard for AI apps |
| How it relates to an LLM | An LLM is one kind | Many run on an LLM as their engine | Lets an LLM app reach live data, such as hotel rates |
Frequently asked questions
What is the primary function of a large language model?
What is a large language model primarily used for?
What is an example of a large language model?
What is fine tuning a large language model?
Can a large language model give wrong answers?
Related terms
See all termsWhere Jetstream fits
An LLM answers from the information it can reach, so the rooms, rates and availability it finds for a property need to be right wherever it looks. Jetstream builds an AI channel manager that keeps a property's rooms, rates and availability accurate on every channel it sells through, managed from one place.