What is a foundation model?
A foundation model is a large AI model trained on a broad sweep of data so it can serve as the base for many different tasks. Rather than training a separate model for translation, another for summarising, and another for writing code, you start from one general model and adapt it. GPT, Gemini, and Claude are foundation models, and so are large image models like the ones behind Midjourney.
The name captures the idea: it is the foundation you build on. The model learns broad, reusable patterns once, at great expense, and then countless products sit on top of it. You reach the specific behaviour you need through prompting, fine-tuning, or connecting the model to your own data, not by training from scratch.
In plain words
Think of a foundation model as a graduate with a broad general education. They have read widely and can hold a conversation on almost anything, but they are not yet an expert in your company. A short onboarding (a good prompt or some fine-tuning) turns that general knowledge into someone useful for your specific job. The expensive part, the years of reading, is already done.
Why it matters
- You skip the hardest, priciest step. Training a model from zero costs millions. Building on a foundation model lets a small team ship AI features in weeks.
- One base, many uses. The same model can draft emails, answer support questions, and help write code. You adapt it per task instead of starting over.
- It set off the current AI wave. Most AI tools you have heard of are a foundation model with a product wrapped around it.
Common pitfalls
- General does not mean expert. Out of the box, a foundation model knows a little about everything and nothing about your business. It needs your context to be genuinely useful.
- You depend on the provider. Building on someone else's model means their pricing, limits, and policy changes become yours too. Plan for that.
- It carries its training in, biases included. A foundation model reflects the data it learned from. Test it on your real cases before you trust it.
Related articles:
- What is an LLM? - The most common kind of foundation model: one trained on language.
- What is fine-tuning? - One way to adapt a foundation model to your specific task.
- What is an AI model? - The broader term, and where foundation models fit within it.
Want to stay one step ahead?
Don't miss our best insights. No spam, just practical analyses, invitations to exclusive events, and podcast summaries delivered straight to your inbox.