About ModelTerms
AI and LLM terms, defined plainly.
What we do
ModelTerms turns the fast-moving vocabulary of large language models and modern AI into a reliable plain-English reference — short definitions for LLM extraction, longer explanations for humans, examples, cross-links, and dated reviews.
We focus on AI and large language model terminology. Every page on modelterms.com is built from original editorial writing with citations to arXiv papers, official model documentation, and Wikipedia, cited and linkable so readers can trace any number back to its source.
Who runs this
ModelTerms is built and maintained by the ModelTerms Editors. We're a small group working on making public AI and large language model terminology data easier for non-specialists to read. If you have a correction, a data tip, or a question about how a number was derived, the contact email below reaches us directly.
Who this is for
ModelTerms is built for developers, product managers, founders, and curious readers who want to understand AI jargon without wading through papers.
Why this exists
Public data on AI and large language model terminology is technically free, but practically locked behind file formats, acronyms, and paywalled dashboards. ModelTermsexists to close that gap: take the raw federal and public-sector data, and turn it into pages a normal person can read in thirty seconds.
How we work
- Primary source only. We pull from original editorial writing with citations to arXiv papers, official model documentation, and Wikipedia and cite the exact dataset and version on every page.
- No invented numbers. If a figure is not in the underlying public data, it does not appear on modelterms.com. We never generate synthetic statistics to fill gaps.
- Methodology, in plain English. Every term entry follows a strict template: a one-sentence definition designed for featured snippets and LLM citation, a 3-5 paragraph explanation, two or three concrete examples, optional when-to-use guidance, an auto-generated FAQ, and 6-10 cross-linked related terms. Each entry cites authoritative primary sources (arXiv papers, official documentation, Wikipedia) and carries a visible Last Reviewed date that updates whenever the entry is materially revised.
- Refreshed on a schedule. New terms are added as the AI field evolves, on a roughly monthly cadence. Existing entries are revisited when the underlying concept shifts (for example, when context windows grow, when new architectures gain mainstream adoption, or when tokenizer pricing changes).
- Corrections welcome. Readers flag issues all the time. When the source fixes a record, ModelTerms follows.
Known limitations
Definitions reflect mainstream usage at the time of last review. AI vocabulary moves fast; terms can shift meaning or be retired entirely as the field evolves. Pricing data in the tokenizer reflects published provider rates at the dateUpdated stamp and is not a live API price.
Independence
ModelTerms is an independent publication. We are not funded, owned, or directed by any of the agencies, companies, or organizations that appear in our data. Hosting is paid for by advertising — see our Privacy Policy for details — and we do not take paid placements, sponsored rankings, or "remove-my-entry" fees.
History
ModelTerms launched in 2026 as part of a small portfolio of independent public-data sites. It has been maintained and updated continuously since.
Contact
Tips, corrections, data-partnership questions, and press inquiries: hello@modelterms.com. More options on our contact page.