Foundation model: A foundation model is a large AI model trained on broad, general data that can be adapted to many different tasks, rather than built for a single purpose. Most of today's chat assistants, image generators, and coding tools are built on one.
Before foundation models, each AI application needed its own model trained for its own job: one for spam filtering, another for translation, another for reading X-rays. A foundation model flips that. It is trained once, at enormous cost, on a huge general corpus, and then adapted for many uses through prompting, fine-tuning, or connecting it to your own data. GPT, Claude, Gemini, and Llama are all foundation models; the chat products you use are applications built on top of them.
The term was coined by researchers at Stanford in 2021 to capture this shift, and it is broader than 'large language model.' A foundation model may handle text, images, audio, and video, and it may power products that look nothing like a chatbot. Your email client's summarizer, your CRM's lead scoring, and a radiology tool may all run on the same underlying model.
The flip side of one model doing everything is that its flaws travel everywhere. A bias or a blind spot in the foundation shows up in every product built on it, which is why vendors publish model cards and why careful buyers ask which foundation model sits under a tool before trusting it with sensitive work.
Example at work
A logistics company evaluates three freight-quoting tools. All three market themselves differently, but their sales engineers confirm each one calls the same foundation model through an API. The real differences turn out to be in how each tool grounds the model in the company's rate data, not in the intelligence underneath.
Why it matters
Most 'AI products' are a thin layer on a handful of foundation models. Knowing this helps you cut through marketing, ask the right questions about data handling, and understand why a limitation in one tool tends to show up in its competitors too.
Related terms
- Large language model (LLM)A large language model (LLM) is a neural network trained on enormous amounts of text to predict the next word, then tuned to follow instructions and hold a conversation. LLMs power ChatGPT, Claude, Gemini, and Microsoft Copilot.
- Pre-trainingPre-training is the first and largest phase of building an AI model, in which it learns general patterns from a massive dataset, such as predicting the next word across trillions of words of text. Later phases then shape it into a useful assistant.
- Fine-tuningFine-tuning is extra training that adjusts an existing AI model's weights using a smaller, targeted set of examples so it performs better on a specific task, style, or domain. It changes the model itself, unlike prompting, which only changes the instructions.
- Frontier modelA frontier model is one of the most capable AI models available at a given time, typically the newest flagship from a major lab. The term is used in policy and industry to describe models whose capabilities may create new risks that need extra scrutiny.
- Multimodal AIMultimodal AI is a model or system that can understand and produce more than one type of content, such as text, images, audio, and video. Modern assistants that can read a screenshot, describe a chart, or listen to speech are multimodal.