A foundation model is one of the handful of big, general-purpose AI models — Claude, GPT, Gemini — trained at enormous scale to do a little of everything. The name captures the idea: it is the foundation other software is built on. Most tools in this directory don't train their own AI; they sit on top of a foundation model and add a workflow, a dataset, or a domain.
"Frontier model" is the same idea with a competitive edge — it refers to the most capable foundation models at any given moment, the ones pushing the frontier.
Why it matters at your desk. Knowing this reframes how you evaluate tools. When you compare two legal or marketing AI products, you're often comparing the same underlying foundation model wrapped in different workflows — so the real questions become which one fits your process, handles your data responsibly, and stays current. It also explains why a model release like Opus 4.8 or GPT-5.5 ripples through dozens of products at once: upgrade the foundation, and everything built on it can improve overnight. Tools like Claude Projects and Perplexity are, at heart, foundation models plus a purpose.
What to watch for: "powered by [famous model]" is a starting point, not a guarantee of quality — the wrapper, the guardrails, and the data it's given still decide whether the result is useful.