LLM API cost calculator. Every model, one workload.

Multiply the tokens you send by the input price and the tokens you get back by the output price, both per 1M tokens, then by the number of requests. Input read from the prompt cache is billed at the lower cached rate, and some models bill long prompts at a higher tier. The calculator does all of that for every model at once.

A piece of a word. In English, 1,000 tokens is roughly 750 words, but each lab's tokenizer splits text a little differently, and images, PDFs and audio count as tokens too. Your API responses report the exact numbers in their usage field.

For a chatbot of 50,000 requests a month (1,500 tokens in, 300 out), the cheapest chat model on Zurelay right now is DeepSeek V4 Flash 0731, at $2.84 a month. Change the workload above to see how the ranking moves: output-heavy and cache-heavy work rank models differently.

Generating a token takes far more compute than reading one, so labs price output several times higher. Capping max_tokens and asking for short answers usually saves more than anything else.

From each lab's published API pricing. Cached input in the list column is discounted by the same ratio Zurelay's cache price uses, and long prompts move to the lab's higher tier where it has one.