Context window
The context window is the amount of text, counted in tokens, that a model can take into account at one time.
Your question, any documents you paste in and the conversation so far all share this space, and in many models the answer does too. When a conversation grows past the window, older parts must be dropped or summarised.
Why it matters: A bigger window lets you work with long documents or large code bases in one go, but filling it costs more and answers can get less reliable on very long inputs.
Related: Token, Maximum output tokensSee the Context window leaderboard
Maximum output tokens
The most tokens a model can write in a single answer, which is often much smaller than its context window.
A model may accept a very long input yet be limited to a shorter reply. Long reports or large code files can hit this limit.
Why it matters: If you need long outputs, check this number as well as the context window.
Related: Context window, Token
API
An API lets software send text to a model and get an answer back, and is how developers pay per token instead of using a chat app.
Chat apps and subscriptions have their own plans and limits. API prices are what the rankings here use, because they are public and comparable.
Why it matters: Prices on this site are API list prices. If you use a chat app you pay a subscription instead, so costs differ.
Related: Token
Open weights
An open-weights model is one whose trained parameters can be downloaded, so you can run it on your own hardware or with other hosts.
This is not always the same as open source. Licences differ: some allow almost any use, others limit commercial use or the size of company. Training data and code are often not shared.
Why it matters: Open weights give you more control and privacy options and often lower prices from competing hosts, but you or a host must run it.
Related: ParametersSee the Open-weights models leaderboard
Parameters
Parameters are the numbers inside a model that are adjusted during training. More parameters usually means a larger, more expensive model to run.
Size alone does not decide quality. Training method and data matter a lot, and many companies do not publish parameter counts for their closed models.
Why it matters: For open-weights models the parameter count tells you roughly what hardware you need.
Related: Open weights, Mixture of experts
Mixture of experts
A design where a model has many specialist sections and only uses a few of them for each token, which saves compute.
A mixture-of-experts model can have a very large total size but use only part of it for each word it writes. That is why such models can be large yet reasonably cheap to run.
Why it matters: It helps explain why a very large model can still be priced low.
Related: Parameters
Reasoning model
A model that works through a problem in extra steps before giving its final answer, usually giving better results on hard tasks.
The extra thinking uses more tokens and time, so reasoning answers cost more and arrive slower. Many models let you choose how much effort to use, which is why one model can appear at several effort levels in some rankings.
Why it matters: For hard maths, coding or planning, reasoning helps. For quick chat or simple rewrites it is usually not worth the extra cost.
Related: Token, Input and output price
Multimodal
A multimodal model can work with more than text, for example reading images, listening to audio or reading PDFs.
Which kinds of input a model accepts varies. Our tables list a feature only when the price list we use states it.
Why it matters: If you need to analyse screenshots, charts or scanned documents, pick a model that accepts images or PDFs.
Related: Context window
Hallucination
When a model states something false or invented as if it were true.
Hallucinations happen even in the best models. They are more likely on obscure facts, recent events and questions where the model should say it does not know.
Why it matters: Check important facts, names, numbers and citations from any model, whatever its ranking.
Related: SimpleQA