The vocabulary behind LLM bills: what each term means, why it changes cost, and which calculator or guide goes deeper.
Asynchronous bulk inference at a discount, usually with higher latency.
A single cost derived from a mix of input and output tokens.
The share of input tokens served from the prompt cache.
The maximum number of tokens a model can consider at once.
Training a model on your own examples to specialise behaviour.
Tokens sent to the model: system prompt, context, and user message.
Sending each request to the cheapest model that can handle it.
Tokens generated by the model.
The standard unit for comparing LLM API prices.
Reusing repeated input tokens at a discounted cache-read rate.
Removing filler, duplication, and low-value context from prompts.
Retrieving relevant documents and adding them to the prompt.
Hidden or visible tokens a reasoning model spends before answering.
Output tokens generated per second.
How long a user waits before the first streamed token arrives.
The unit AI providers bill for.
A planned number of tokens for a workload or feature.