What will your AI actually cost per month?

Input tokens are what you send to the model (your prompt, conversation history, documents). Output tokens are what the model generates back. Output is almost always priced several times higher than input — often 5-8x — because generating text is more computationally expensive than reading it.

For providers that support it, a cached input token typically costs 10-20% of the normal input price — some providers go much lower for high-volume caching. If your app resends a large, unchanged prefix (a system prompt, tool definitions, a knowledge base) on every request, caching is usually the single biggest cost lever available.

It depends entirely on the specific model tier, not the provider name. Each provider offers a range from budget models (well under $1/M tokens) to flagship models ($5-50/M tokens). Use the calculator above with your actual usage numbers — the cheapest model for a simple classification task is rarely the cheapest model for a complex agentic workflow.

DeepSeek's models are priced far below the major US providers, which has made them a common budget option for high-volume, latency-tolerant workloads. As with any pricing comparison, cost isn't the only factor — evaluate quality and reliability for your specific task before switching purely on price.

Often enough that this page shows a verification date, not a promise. Every major provider has adjusted pricing multiple times within a single year — sometimes lowering it as models get more efficient, sometimes introducing new tiers. Always confirm current rates on the provider's own pricing page before committing a budget for a long-running project.