The model landscape: GPT, Gemini, Claude, Llama and more
Models, apps and companies are easy to mix up. Learn who makes the main model families, the difference between open and closed models, and how to choose without chasing every new release.
News about AI is full of names: GPT, Gemini, Claude, Llama, DeepSeek, Mistral, Qwen. They are not all the same kind of thing, and new versions arrive every few months. This lesson gives you a stable map. It names model families and who makes them, but avoids version numbers and rankings, because those go out of date quickly.
Model, app, company: three different things
Untangling the names
| Layer | What it is | Example |
|---|---|---|
| Company | The lab that trains the model | OpenAI, Google, Anthropic, Meta |
| Model (family) | The trained neural network, released in versions and sizes | GPT, Gemini, Claude, Llama |
| App | The product you chat with, which wraps a model with search, file upload, memory and so on | ChatGPT, the Gemini app, the Claude app, Meta AI |
| API | How developers call the model from their own code | OpenAI API, Gemini API, Claude API |
So "ChatGPT" is an app; GPT is a family of models inside it. Two apps can use the same model and still feel different, because the app adds its own instructions and tools. A model's name also usually includes a version and a size or tier, and the app may pick the model for you.
The main families
Who makes what (as of 2026; check each provider for the latest)
| Family | Made by | Notes for learners |
|---|---|---|
| GPT | OpenAI | Powers ChatGPT. Also available through an API. |
| Gemini | Powers the Gemini app and features across Google products. Google AI Studio gives developers a free tier with rate limits. | |
| Claude | Anthropic | Powers the Claude app and API. Often used for writing, long documents and coding. |
| Llama | Meta | Released with downloadable weights under Meta's licence. Used in Meta AI and by many developers. |
| Mistral | Mistral AI (France) | Offers both open-weight and hosted models. |
| DeepSeek and Qwen | DeepSeek and Alibaba (China) | Several models with openly downloadable weights. |
| Copilot | Microsoft | An assistant product built into Windows, Edge and Microsoft 365, using models from several providers. |
This is not a complete list, and the competitive order shifts often. Each family also comes in sizes: small, fast and cheap models for simple tasks, and larger, slower, more expensive ones for harder reasoning.
Open versus closed models
Two ways models are released
| Question | Closed (hosted) models | Open-weight models |
|---|---|---|
| How you use it | Through the company's app or API | Download the weights and run them yourself, or use a host |
| Examples | GPT, Gemini, Claude | Llama, Mistral's open models, Qwen, DeepSeek |
| Cost to start | Free tier or subscription; API charged by tokens | Free to download; you pay for the computer |
| Privacy | Your text goes to the provider's servers | Can run on your own laptop or server, so data stays with you |
| Control | The provider can change or retire models | You can keep the exact version and fine-tune it |
| Effort | Very low | Needs hardware and some technical skill |
"Open" usually means open weights (you can download the trained numbers), not necessarily open source in the strict sense. The licence decides what you may do with them, including commercial use, so read it before building a product.
Running a model on your own laptop
Tools like Ollama and LM Studio let you run smaller open models locally. Small models run on an ordinary laptop with enough memory but are weaker than the big hosted ones. They are useful when privacy matters or when you want to learn how things work without an internet connection. The *Free AI tools you can use today* lesson shows where to start.
Kinds of models you will meet
Not every model is a text chatbot
| Kind | What it does | Example use |
|---|---|---|
| Chat / text (LLM) | Reads and writes text | Study help, emails, code |
| Reasoning models | Spend extra effort thinking step by step before answering | Hard maths, logic, tricky bugs; slower |
| Multimodal | Accepts images, audio or documents as well as text | Photograph a worksheet and ask for help |
| Image and video generators | Create pictures or clips from text | Posters, thumbnails |
| Embedding models | Turn text into numbers so similar meanings can be found | Search over your own documents |
| Speech models | Convert speech to text and back | Voice assistants, transcription |
How to choose without chasing hype
A simple way to pick
- 1Start with the job. Quick question, writing, study help, research with sources, working with a document, code, an image? Different tools suit different jobs.
- 2Use the free tier of two tools on the same real task and compare. Your own tasks are a better test than any chart on social media.
- 3Check what you need around the model: web search, file upload, voice, image input, how long chats can be, whether it handles your language well.
- 4Think about your data. If it is private or belongs to your employer or college, follow the rules and avoid pasting it (see *Privacy: what never to paste into an AI tool*).
- 5Re-check now and then. Models improve fast. Revisit your choice every few months rather than every week.
Beware of leaderboards and "best model" posts
Benchmark scores and rankings are useful signals but narrow. A model that tops a coding test may not be the best for your essay or your language, and companies can tune for popular tests. Treat rankings as a hint and trust your own comparison on your own tasks.
I want to compare AI tools for this task: [describe your task, e.g. summarise my Operating Systems notes into 10 revision flashcards].
Here is the input:
---
[paste a short sample]
---
Do the task. Keep the answer under [250] words.What to expect: Paste exactly the same prompt into two tools. Score each answer from 1 to 5 for accuracy (against your notes), clarity and usefulness, and keep a note of which one you prefer for this kind of task.
How this comes up in interviews
Common questions: "Which AI tools do you use and for what?", "What is the difference between open and closed models?" and "How would you choose a model for a project?". Strong answers give a concrete task, name what you compared (quality, cost, privacy, speed, language support) and admit that the answer changes as models change.
Try it: Build your own model cheat-sheet
Make a small table with one row per tool you try (for example ChatGPT, Gemini, Claude, Microsoft Copilot). Columns: maker, free tier available (yes/no and what limits you hit), web search (yes/no), file upload (yes/no), best for, one weakness you noticed. Fill it in after using each tool for at least one real task.
Key takeaways
- Company, model, app and API are different layers: ChatGPT is an app, GPT is the model family inside it.
- Main families: GPT (OpenAI), Gemini (Google), Claude (Anthropic), Llama (Meta), plus Mistral, DeepSeek, Qwen and others.
- Closed models are used through the provider; open-weight models can be downloaded and run yourself, subject to their licence.
- Models come in sizes and kinds: reasoning, multimodal, image, embedding and speech models.
- Choose by task, test free tiers on your own work, and treat leaderboards as hints only.
Check yourself
Finishing the quiz marks the lesson as done.
1. What is the relationship between ChatGPT and GPT?
2. What does "open-weight" mean?
3. A college project handles student records and you want the text to stay on your own machine. Which option fits best?
4. A post says one model is "the best" because it tops a benchmark. What is the wisest response?
5. Which kind of model turns text into numbers so that similar meanings can be matched in search?
Saved on this device. Create a free account to keep it on every device.
Practise this with Sam
A real mock interview, out loud, with an honest scorecard. Free to start.