Compare AI models side by side

How to use the comparison tool

Type a few letters of a model’s name in the search box and the list filters instantly. Add 2 to 4 models and their specs line up row by row for easy reading. For each criterion, an automatic verdict flags the model that comes out ahead (cheapest, largest context window, best score, and so on). The share link lets you send your comparison to a colleague or come back to it later.

What each row means

Token. The basic unit a model reads and writes: a fragment of text, often part of a word. Everything is counted in tokens, from pricing to text length. To see how many tokens your text represents, try the token estimator.

Input and output price. Both are shown per million tokens. Input is what you send to the model (your question, a document, the conversation history); output is what it sends back. Output is usually the more expensive of the two.

Context window. The model’s “working memory”: the maximum amount of text it can take into account at once, prompt and documents included. A large window lets you analyze a long report in one go. The context window tool helps you picture what that means.

Max output. The maximum length of a single answer. It is often much shorter than the context window: a model may read a very long document yet only write a reply of limited length.

Features. Tools: the model can call external functions (search, calculations, software), which is essential for building an agent. Reasoning: it can “think” step by step before answering, useful for complex problems. Vision: it understands images. Audio: it handles sound or voice.

Open weights. Weights are the parameters a model learns during training. When they are published, you can download the model and run it on your own servers, which is a game changer for privacy. Otherwise, the model is only available through its maker’s service.

Release date. Handy for spotting recent models, but an older model is often still a great choice if it is cheap and good enough for the job.

Benchmark scores. Results on standardized tests, drawn from the same data as our AI ranking. An empty cell means the model was not evaluated on that test, not that it failed.

How to choose for your use case

• Writing and everyday questions: a budget model is usually enough. Focus on the output price.
• Long document analysis: favor a large context window and check the input price, since that is what adds up when you send lots of text.
• Agents and automations: make sure the tools feature is there, then estimate the bill with the AI agent cost calculator, because an agent makes many calls in a row.
• Sensitive data: an open-weight model, hosted by you or a provider you trust, limits where your data travels. See our pick of open source AI models.
• Hard problems (code, math, logic): the reasoning feature and benchmark scores move to the top of the list.

Common pitfalls

• Only looking at the input price. A model that is cheap on input can get expensive if your answers are long.
• Forgetting reasoning tokens. Models that “think” often generate intermediate tokens, usually billed as output. Your real bill can end up higher than your estimate.
• Assuming a huge context is always fully used. Quality can drop when the window is filled to the brim; send only what is relevant.
• Treating the automatic verdict as advice. It compares figures row by row without knowing your use case. Weighing the criteria is up to you.
• Mixing up models and assistants. If you are looking for a personal subscription, the AI assistants comparison is a better fit.