AI model comparison: prices, context windows and features of 340+ LLMs

Choosing a model usually starts with three questions: can it handle my input, can it do what I need, and what will it cost? This table answers all three for 340+ models at once, so you can narrow the field before testing anything.

Reading the prices

Prices are list prices in US dollars per million tokens. Input covers everything you send (system prompt, history, documents); output is the model’s reply and costs more because it’s generated one token at a time. Cached input is the discounted price for re-reading a prompt prefix the provider has already cached, which matters for chatbots and agents that resend the same instructions on every request. A dash means the provider hasn’t published that price.

Context window, max output and the cost of filling it

The context window is the most a single request can hold, prompt and reply together. Max output limits the reply on its own. The Fill context column multiplies the window by the input price, which tells you what one maximum-length prompt costs. Two models can share a 1M-token window and still differ by more than a hundred times on this column, so it’s a quick way to see which long-context models are practical for everyday use.

Capabilities and open weights

The capability filters (image input, tool calling, reasoning, prompt caching, structured output) use the features each provider declares for its API. They tell you a feature exists, not how well it works. Open-weight models publish their weights, so several companies host them and you can also run them yourself; their price here is a typical hosted price, and our VRAM calculator covers self-hosting.

Where the data comes from

The table is generated from the OpenRouter models API and the open-source LiteLLM price list, refreshed every day. We correct the data by hand when a source is wrong; for example, when a feed reports a cache price as the normal input price, the model is held back until it’s checked against the provider’s pricing page. Release dates are when a model was first listed by our source, which can differ from the announcement date. The methodology page has the details.