All AI models and their profiles

Model or assistant: what’s the difference?

When you chat with ChatGPT, Claude or Gemini, you are using an assistant: an app with an interface, a history and sometimes a subscription. Behind that assistant runs an AI model, a program trained on huge amounts of text to understand a request and produce an answer. The model is the engine; the assistant is the car.

This distinction matters in practice. The same assistant can switch models over the months, or let you pick between several. And the same model can power many different services, or be used directly by developers through an API (a technical gateway that lets software send requests to the model). This page lists models; for consumer apps, see our AI assistant profiles instead.

Using the filters

You can filter the table above in a few clicks:

• Lab: show only the models from a given company. To learn more about each one, see the AI labs index.
• Price: spot budget models or, on the contrary, premium ones. Prices are shown per million tokens (a token is a small chunk of text).
• Context window: how much text the model can read at once. Useful if you work with long documents.
• Features: tool calling (essential for an agent), reasoning, vision or audio.
• Open or closed: an open model publishes its weights, so you can run it on your own machines; a closed model is only available through its maker’s service. Open source AI models have their own page.

Filters can be combined: for example, you can look for an open model with vision from a European lab.

What a model page contains

Each model page brings together:

• the lab, release date and license type (open weights or not);
• input and output prices, context window and max output;
• supported features;
• benchmark scores when the model has been evaluated, with its position in the AI ranking;
• the source and date of every data point.

To put several models head to head, use the model comparison tool.

Which models get a page?

There are a huge number of models and variants. To keep the index readable, a selection rule applies:

• models from the 33 labs we follow, as listed on models.dev, an open database of AI models;
• released within the last 24 months, or still offered by their maker;
• one page per underlying model: a model offered by several hosting providers appears only once;
• priority to models evaluated by Epoch AI, for which benchmark scores are available.

Models without a page remain available in the comparison tool and in the AI agent cost calculator. The rule is explained in full in our methodology.