The AI models
ChatGPT, Claude, Gemini, Grok and the rest are all built on AI models. Here is what a model is, how the main ones differ, and how to choose between them.
What a model actually is
A model is the trained part of an AI. It is the result of training: a computer shown vast amounts of material until it got good at a task. The chatbot you type into is the app wrapped around a model, and the same company often sells several models at once.
Think of the model as the engine and the app as the car. Two cars can share an engine, and a company can fit a bigger engine to the same car later. That is why an app can get smarter without changing how it looks.
The name shown in the app is usually the model's name. If you are not sure which one you are using, check the settings or the model menu.
Six things that make models different
Most of the differences between models come down to these.
- 01
Size
How many settings it has
Models are measured in parameters, the billions of internal settings they adjust while learning. Bigger usually means more capable and more expensive to run, though not always.
- 02
Speed and depth
Quick answers or careful thinking
Some models answer straight away. Reasoning models take longer to work a problem through. That makes them slower and dearer, but often stronger on maths, logic and planning.
- 03
Working memory
How much it can keep in view
The context window is how much a model can keep in mind at once, including your conversation and any documents you share. A bigger window suits long reports.
- 04
Senses
What it can read beyond text
Some models only read and write text. Multimodal models can also work with pictures, audio and video, which is handy for reading a photo of a receipt or a chart.
- 05
Doing things
Answering versus acting
Some models can only answer. Others can search the web, use your calendar or files, or take a series of steps for you. Those are agent-style tools, and they need more care.
- 06
Open or closed
Whether you can download it
A closed model runs only through its maker's service. An open-weight model can be downloaded, and some are small enough to run on a local computer or phone.
The main families, November 2026
Version names change every few months, so read these as a snapshot.
- 01
Claude
Anthropic
Claude comes in tiers. Fable 5.1 is the top publicly available model, in a tier above Opus. Opus 5.5 is the large, careful option, Sonnet 5.5 balances cost and capability, and Haiku 5.5 is the quick, lighter one. Often chosen for: long documents, careful writing and coding help.
- 02
ChatGPT
OpenAI
ChatGPT runs on OpenAI's GPT models. The GPT-5.6 family, released in July 2026, comes in three tiers, Sol, Terra and Luna, so you can trade quality against speed and cost. Often chosen for: everyday questions, images and a wide set of built-in features.
- 03
Gemini
Google
Gemini is Google's family of models, in sizes from fast and light to large. Gemini 3.5 Flash, a fast model, was released in May 2026. It also powers features in Google's own apps. Often chosen for: searching and working with Google's apps and documents.
- 04
Grok
xAI
- 05
Llama and Meta AI
Meta
- 06
DeepSeek
DeepSeek, China
DeepSeek is known for capable models that are cheap to run and free to download. DeepSeek-V4, an open model, was previewed in April 2026. Before you paste anything personal into any provider you do not know, check where your data is stored. Often chosen for: low cost and open-weight models.
- 07
Mistral
Mistral AI, France
Mistral AI is known for efficient models, many of them open-weight and free to download. Often chosen for: efficient models and a European provider.
Which kind of model for which job?
Start with the job, then look for the feature that matters for it.
| The job | What to look for |
|---|---|
| Reading or summarising a long report | A large context window. Check the limit in the app, as paid plans often allow more. |
| Research you can check | A search or deep research mode that shows its sources. Open the links before you rely on them. |
| Writing and editing | Almost any current model will do. Read the output and rewrite it in your own voice. |
| Maths, logic or planning | A reasoning model. Slower, but it works through problems more carefully. |
| Code | A coding agent or a reasoning model. Always run and test what it writes. |
| Pictures and video | A multimodal model, or a dedicated image generator. |
| Keeping costs down | A smaller model or a free tier. Check the usage limits before a busy week. |
| Keeping data on your own device | A local model. Usually more private, often less powerful. |
Test two models yourself
A five-minute test will tell you more than any league table.
- 01
Pick a real task
Choose something you actually need from your week, not a trick question.
- 02
Ask two assistants
Give the same wording to both. Use the free versions if that is all you have.
- 03
Check one fact
Compare one claim in each answer with a source you trust, and ask each assistant where its information came from. Watch out for confident answers that turn out wrong, which is called a hallucination.
- 04
Keep the better one
Keep the answer you would use with fewer edits. That is the right model for you, for now.
Three things to check in any app
- 01
Which model am I using?
The name shown in the app is the model. Free and paid plans can use different models, so check before you compare.
- 02
How recent is it?
Every model has a knowledge cutoff. Without a search tool, it will not know about anything after that date.
- 03
What happens to my data?
Check whether your chats are used for training and how to turn that off. See data privacy.
Why this page will date
The big labs release new versions every few months, and the rankings change with them. Where a model name is not yet settled, we describe the family rather than guess at the version. We look at this page alongside each fortnightly round-up.
- BenchmarkList, model release timeline (GPT-5.6 family, Gemini 3.5 Flash, DeepSeek-V4)
- AIFOD Realtime, AI model releases, October 2026
- PromptZone, AI model releases timeline (Claude 5 family)
- Anthropic (Claude model names and tiers)