The basic pricing unit is usually tokens
Most text-generation APIs charge for tokens, not full words or requests. A token is a small chunk of text. Your prompt, system instructions, pasted context, and the model's reply all contribute to the final cost.
That is why a usage estimate should include both request volume and token volume. A low request count can still become expensive when each request carries long documents or detailed generated answers.
Input, output, and cached input prices
- Input tokens are the text sent to the model, including instructions, user messages, and context.
- Output tokens are the text generated by the model. These often cost more because generation uses more inference work.
- Cached input tokens may apply when a provider can reuse repeated prompt or context segments at a lower price.
Why prices differ by provider and model
Providers price models differently based on capability, latency, context window, modality, and infrastructure cost. A premium reasoning model can be the right choice for complex work, while a smaller model may be better for classification, routing, or short support replies.
Use CostRivo's provider pages and LLM API pricing table to compare model prices before turning them into monthly estimates.
Estimate before launch
The safest time to model API cost is before traffic arrives. Estimate a realistic average request, then test high-usage cases such as long chat histories, retrieval context, or verbose responses.
After you have a baseline, compare alternatives such as OpenAI vs Anthropic pricing to see how the same workload changes across providers.