"Foundation model" is a term you'll see in vendor pitches, analyst reports, and procurement reviews. Here's what it actually means and why it matters when you're evaluating AI tools.
A foundation model is a large AI model — trained on broad data, capable of many tasks — that other AI applications build on top of. Examples: Claude (Anthropic), GPT-5 (OpenAI), Gemini (Google), Llama (Meta). Most AI vendor products are wrappers around one or more foundation models.
Before 2022, AI models were usually narrow — one model for translation, another for image classification, another for chess. Foundation models are different: trained on broad data, capable of many tasks, adaptable to new uses with minimal additional training.
The term was coined by Stanford in 2021 to describe this new class of model. By 2026, it's standard vocabulary in AI procurement.
When evaluating an AI vendor, one critical question is: which foundation model does this product use? A vendor product is only as good as the model underneath it.
Some vendors disclose openly ("powered by Claude 3.5 Sonnet"). Others obscure it. Strong vendors disclose; weak ones often do not.
If you cannot get a clear answer about the foundation model, you have either a thin wrapper hiding behind branding, or a custom model the vendor is overselling. Both are yellow flags.
Closely related. LLMs (large language models) are the most common type of foundation model. Some foundation models also handle images, audio, video — making them "multimodal" foundation models.
Yes. The foundation model determines quality, cost, and capability. A current frontier model (Claude Sonnet 4.7, GPT-5) under the hood is meaningfully different from an older or smaller model.
Sometimes — some vendor products let you pick. Often not. Worth asking before buying.
Increasingly yes for frontier capability — Claude, GPT, and Gemini are all excellent. Differentiation is moving to deployment, integration, and application layers.
Yes, on a roughly 6-12 month cadence. Worth re-evaluating capability assumptions every 6-12 months.