Skip to content
BENIBIX Learn

AI Labs and Model Families

Level 1 · The AI labs

What an AI lab is, and how a lab, a model family and an app differ. Who makes GPT, Claude, Gemini, Llama and Qwen.

Course contentsLesson 1 of 13

After this lesson you can

  • Tell a lab, a model family and an app apart
  • Name the lab behind five model families
  • Say why most AI apps are not labs

10 mincovers 5 glossary entries

What an AI lab is, and what it makes

An AI lab is a company that trains its own large AI models. OpenAI, founded in 2015, makes the GPT models. Anthropic, founded in 2021, makes Claude. Google DeepMind makes Gemini. Each lab gives its models a family name, and each new version keeps it. Most AI apps you meet run on a model from one of a few labs.

Three names get mixed up all the time. The lab is the company that built the model. The model family is the brain, such as GPT or Claude. The app is the window you type into, such as ChatGPT. OpenAI is the lab, GPT is the family, and ChatGPT is the app.

A model is a large file of numbers, called weights or parameters. The lab sets those numbers by training on a huge pile of text, which is the costly part. A large language model, or LLM, is the kind that reads and writes text. Once the file exists, it can be copied and run for millions of people at once.

A lab shares a model in one of two ways. A closed model stays on the lab's own servers, and you reach it over the internet. An open-weight model is a file you may download and run yourself. Some labs do both. The LLM course covers what each choice means for you.

Most AI companies are not labs at all. They build an app on a model that a lab trained, and pay the lab for each use. A writing app, a help desk bot and a coding tool can all run on the same model behind the scenes. When the lab changes the model, all of them change.

The labs sit in a few places. The best known are in the US: OpenAI, Anthropic, Google DeepMind, Meta and Microsoft. China has DeepSeek, Alibaba's Qwen team, Moonshot AI, Z.ai and MiniMax. Mistral AI in Paris is the best known in Europe. Many of the Chinese labs share their weights.

People often use one brand name for all of AI. GPT is not a word for every chatbot, and Claude is not a kind of GPT. They are rival families from rival labs. Knowing who made a model tells you where your words go, what it costs and what rules come with it.

Who makes which model family

LabModel familyOpen or closed
OpenAIGPTMostly closed
AnthropicClaudeClosed
Google DeepMindGemini, and GemmaGemini closed, Gemma open
MetaLlamaOpen weights
Mistral AIMistral and MixtralSome open, some closed
DeepSeekDeepSeekOpen weights
Alibaba's Qwen teamQwenMostly open weights
A lab trains the model. Most of the apps you use are built by someone else on top of it.

Worked example

Emma finds the model behind her app

Emma pays $10 a month for a writing app that says only 'powered by AI'. Its help page names the model: a Claude model from Anthropic.

Her friend's notes app, at $6 a month, runs on the same model. That is why both apps write in such a similar way.

The words for this lesson5

Check yourself

What is the difference between ChatGPT and GPT?

Finished this lesson?

Marking it complete also marks its 5 glossary entries learned, which updates your roadmap and your saved page. 0 are already marked.