learn / ai for everyone / ae2 · 1 of 1 · 8 min
Claude, ChatGPT and Codex, Gemini, Grok, DeepSeek, Qwen, GLM and the rest. Who builds them, what each is known for, and what "open weights" means for you.
## Company, model, app
People say "I asked ChatGPT" the way they say "I Googled it", but there are three different things in play:
company model family app you actually use ───────── ──────────── ──────────────────── Anthropic ──▶ Claude ──▶ claude.ai, Claude Code OpenAI ──▶ GPT ──▶ ChatGPT, Codex Google ──▶ Gemini ──▶ Gemini app, Gmail, Docs
A model is a large language model (LLM): a huge file of learned numbers that predicts text. Each company releases a family of models, usually a big, slower, smarter one and smaller, faster, cheaper ones, and replaces them every few months. Apps wrap a model with extras: web search, file uploads, memory, tools. When two apps "use the same model", the extras and hidden instructions still make them behave differently.
Your coworker says "Copilot and ChatGPT gave me different answers, so they must be totally different AIs." What's the best response?
## The big closed labs
These companies keep their model files private. You use them through their apps or pay to call them from software.
- Claude (Anthropic). Known for careful writing, long documents and coding. Claude Code is its coding agent that works inside your project.
- ChatGPT and GPT (OpenAI). The best-known chat app, with images, voice and a huge plugin ecosystem. Codex is OpenAI's coding agent. Microsoft Copilot is built largely on OpenAI models.
- Gemini (Google). Built into Gmail, Docs, Android and Search. Strong with video, images and very long inputs.
- Grok (xAI). Built into X (formerly Twitter), with live access to posts there. Marketed as having fewer guardrails.
None of these is "the best" at everything, and the leader changes every few months. That's why this lesson has a review date.
## Open-weight models
Some labs publish the model file itself, so anyone can download it, run it on their own hardware, or host it for others. These are called open-weight models.
- DeepSeek. A Chinese lab whose models rival the closed labs at a fraction of the price.
- Qwen (Alibaba). A large family, from tiny models that run on a laptop to very large ones. Popular for local use.
- GLM (Zhipu AI, also called Z.ai). Strong at coding and agent work, and popular as a cheap engine inside coding tools.
- Others you'll see: Llama (Meta), Mistral (France), Kimi (Moonshot AI), Gemma (Google's small open models).
Open weights let you run a model locally (covered in Running AI on Your Own Computer), pick a cheaper host, or fine-tune it. Small open models run on a decent laptop. The biggest ones need serious graphics cards (GPUs) or a cloud provider. Open weights isn't the same as open source: the training data and code usually stay private, and licenses vary.
What does it mean that Qwen and DeepSeek release "open-weight" models?
## Where do your words go?
This matters more than which model is smartest.
- Using a company's app or website means your messages are stored on that company's servers, under the laws of the country it operates in. That's true of every company in this lesson.
- Some apps hosted in China (for example the official DeepSeek, Qwen and GLM apps) fall under Chinese law, which can require companies to share data with the government. Many employers and governments ban them on work devices for that reason.
- The same open model run on your own computer, or by a host you trust, sends nothing to the lab that made it.
The rule: judge the app and the host, not just the model. Check the privacy settings for whether your chats are used for training, and never paste in things you'd regret leaking. More in Privacy: What Not to Paste.
Your company bans a chatbot app on work laptops, but allows the same open-weight model running on the company's own servers. Why does that make sense?
## Big, small and "thinking"
Inside each family you'll usually see three kinds of choice:
- Big vs. small. Big models are smarter and slower and cost more. Small ones are quick and cheap, and fine for simple jobs.
- Thinking or reasoning modes. The model works through the problem before answering. It's slower, but much better at math, logic, planning and code.
- Free vs. paid plans. Paid plans usually get the bigger model, more messages, longer files and agent features.
A good default: use the regular model for everyday questions, switch to thinking mode when it gets something wrong or the task has many steps, and use a small fast model for bulk, simple work. Which Model for Which Job goes deeper.
## A starter cheat sheet
- Writing, editing and long documents: Claude, ChatGPT or Gemini all do well, so try two and keep the one whose style you like.
- Your life already runs on Google: Gemini, because it can see your Gmail, Docs and Calendar if you let it.
- What people are saying right now: Grok for posts on X, or any assistant with web search turned on.
- Coding: Claude Code, Codex, or an open tool like OpenCode using whichever model you prefer.
- Private or offline: an open-weight model like Qwen or Gemma running on your own machine.
- Cheap at scale: open-weight models such as DeepSeek, Qwen or GLM through a host you trust.
Try the same question in two assistants this week. Noticing how they differ teaches you more than any ranking.