Is ChatGPT an LLM?
Yes — ChatGPT is an interface over large language models, and it is equally correctly called AI, generative AI and a chatbot. Two corrections worth making: the models are now multimodal, learning from images, audio and video as well as text, so "language model" is slightly narrow. And it is not artificial general intelligence — OpenAI's own mission treats AGI as something to be reached, not shipped.
Four labels fit. The fifth is the one worth arguing about.
LLM. OpenAI's description of the mechanism is textbook: the model "analyzes relationships within this data—such as how words typically appear together in context—and uses that understanding to predict the next most likely word", working in tokens that "may represent whole words, parts of words, or punctuation". Large sets of numbers called weights, trained by prediction. That is what the term names, so yes.
The multimodal wrinkle. The same article says the models learn "from large amounts of information, including text, images, audio, and video", and that models generating other kinds of content "learn patterns in how pixels relate to each other". Strictly, calling ChatGPT a language model is a couple of years out of date — the family it belongs to now handles more than language, which is why "foundation model" appears in OpenAI's own article title.
Generative AI, and a chatbot. Yes and yes. Generative is literally the G in GPT — it composes rather than retrieves. Chatbot is accurate for the chat interface and steadily less descriptive of the product, which now also runs multi-step agentic tasks, produces documents and writes code. Nobody calls a thing that opens your terminal and edits a repository a chatbot, but Chat is still one of its three experiences.
AGI: no. OpenAI's stated mission is "to ensure that artificial general intelligence benefits all of humanity" — the grammar of an objective, not a description of inventory. Its own documentation simultaneously states that ChatGPT hallucinates, may express high confidence in wrong answers, and cannot know about events after its training cutoff unless a tool fetches them. A company does not describe general intelligence in those terms.
Why the label changes what you expect.
The most useful consequence of the label is diagnostic. A system that predicts likely continuations will be excellent at things that are patterned — grammar, structure, style, translation, summarising material you supplied — and unreliable at things that require a lookup: a specific date, a real citation, an exact figure.
That is exactly the failure profile OpenAI documents, right down to fabricated references to non-existent sources. Knowing it is an LLM tells you which half of your task is safe, which is more practically useful than any benchmark. We take that apart on the accuracy page.
Precision helps here. GPT-5.6 Luna and Sol are models. ChatGPT is a product that puts an interface, tools, memory, file handling, search and agentic execution around them. The same models are sold raw through the API with none of that.
This matters when you compare things. "ChatGPT versus Claude" usually compares two products; "GPT-5.6 Sol versus Claude Opus" compares two models. The prices are completely different questions too — a subscription buys the product, per-token rates buy the model.
Around 2,400 people a month ask whether ChatGPT is artificial general intelligence, and the question deserves better than a smirk. There is no agreed test for AGI, so nobody can settle it definitively — but the ordinary sense of the term implies reliably transferring competence across arbitrary domains, at human level, without supervision.
The documented behaviour of the current product does not meet that: it fabricates citations, is sensitive to phrasing, cannot know recent events without a tool, and answers the same question differently on different runs. Those are not marginal edge cases; they are in the vendor's own limitations list. No is the honest answer, and it will stay the honest answer until something changes that list.
Some of this cluster is really asking whether the thing is thinking. It is artificial intelligence by any standard use of the term, and OpenAI's own current definition opens with "an artificial intelligence-based service".
Whether it understands anything is a philosophical question this page cannot settle. What the documentation does settle is the mechanism — prediction over learned patterns, with models that OpenAI says do not store copies of their training data but adjust parameters to reflect it. Whatever that is, it is neither a lookup table nor a person, and most confident claims in either direction are overreaching.
Same category, different vendors.
If the question is really "which LLM should I use", these are the three comparisons that answer it. Pricing on each is derived from our snapshot, not retyped from a vendor page.
Frequently asked.
Quick follow-ups people search after this question.