Estimate the monthly cost of using LLMs from the requests per month and the average input and output tokens per request. Useful for budgeting AI API usage.
Filter the table by provider and by model type (reasoning, flagship, balanced, economy), or search for a model by name. Enable Include deprecated to also show retired models.
Click the column headers to sort by input price, output price, context or total monthly cost, and compare models at a glance.
Prices are indicative and updated regularly; the last-updated date appears below the table. Input tokens include the entire conversation history sent with each request, so the real cost grows with longer conversations.
Pick a model, how its weights are stored (fp16, q8, q4), the context and how many conversations run at once. You will see how much memory the weights, the KV cache and a margin need, whether it fits on each GPU, the longest context that fits and the tokens-per-second ceiling set by its memory bandwidth. It is a calculation, not a measurement: a real engine reserves a bit more.
Move the parameters and the training tokens, or pick a well-known model. The curve is the loss the Chinchilla law predicts for every way of spending the same compute; the green dot is the best one. It helps you understand the trade-off, not predict the quality of a specific model.