What does LLM stand for?
The short answer: in AI, LLM stands for Large Language Model — the kind of program behind ChatGPT, Claude and Gemini: a neural network trained on enormous amounts of text until it can predict, word by word, what text should come next. (Heard it in a law-school context? There it’s a Master of Laws degree — unrelated, same initials.)
The three words, unpacked
Large: measured in parameters — the internal numbers the model learned during training. Consumer-runnable models range from ~1 billion to ~70 billion parameters (written 1B to 70B); frontier cloud models go far beyond. Language: trained on text — books, code, the web — so its native skill is producing text, including source code. Model: a mathematical function, not a database. It doesn’t look answers up; it generates them, which is why it can be brilliant and confidently wrong in the same paragraph.
How an LLM works in 60 seconds
Text is split into tokens (word chunks). Given all the tokens so far, the model computes a probability for every possible next token, picks one, appends it, and repeats — thousands of times a second. Everything an LLM does — answering, translating, coding — emerges from that single trick applied at scale. Two numbers describe any model: its parameter count (capability ceiling) and its context window (how many tokens it can consider at once — from ~8K to over 1M on recent models).
The other LLM (for the law students)
Universities also use LLM for the Legum Magister — Master of Laws, a postgraduate law degree. If you searched “LLM” after seeing it on a CV or a law-school site, that’s your answer and the rest of this page isn’t for you. Everywhere in tech, LLM means Large Language Model.
Yes, you can run one on your own computer
This surprises people: open-weight LLMs from Meta (Llama), Google (Gemma), Alibaba (Qwen) and Mistral are free to download and run locally — private, offline, no subscription. A modern gaming GPU or Apple Silicon Mac runs a genuinely useful one today. The easiest on-ramp is Ollama; our configurator tells you which model your exact machine can handle, and the local LLM leaderboard ranks all of them.
Five terms you'll meet next
| Term | Meaning |
|---|---|
| Token | A chunk of text (~¾ of a word on average) — the unit LLMs read and write |
| Parameters (7B, 70B…) | The learned numbers inside the model; more ≈ more capable, more memory |
| Context window | How much text the model can consider at once |
| Quantization | Compressing a model to fit consumer hardware — explained here |
| Inference | Running the model to generate text (vs. training, which builds it) |
Frequently asked questions
Is ChatGPT an LLM?
ChatGPT is an application powered by LLMs (OpenAI's GPT series). The distinction matters: the app adds the chat interface, tools and memory around the underlying language model.
What does the B mean in 7B or 70B?
Billions of parameters — the learned internal values. A 7B model has ~7 billion of them. More parameters generally means more capability but also more memory needed to run it.
Can I run an LLM on a normal computer?
Yes. Quantized open-weight models run on ordinary gaming GPUs and Apple Silicon Macs — an 8GB graphics card already runs an 8-9B model well. Tools like Ollama make it a one-command install.
Is an LLM the same thing as AI?
An LLM is one kind of AI — the kind specialized in language. AI is the umbrella; image generators, recommendation systems and chess engines are other branches.
By Mohamed Meguedmi — independent comparator of locally-runnable LLMs, benchmarked on a real RTX 5070 Ti (data CC BY 4.0). See the local LLM leaderboard and the best Ollama models.