What it costs to run thousands of PDFs through an LLM.
Dumping entire documents into the model is easy and expensive. Retrieve the relevant pages first. For retrieval cost see the RAG chatbot cost calculator.
The PDF Processing with AI Cost Calculator estimates what it costs to run a batch of PDF documents through a large language model each month. It multiplies your PDFs per month by pages per PDF and tokens per page to get total input tokens, then adds the output tokens per PDF the model generates back. Those two token totals are priced separately against your input $/1M and output $/1M rates and summed into a monthly figure. The main cost drivers are document volume, page density, and how much text each page contributes β long or image-heavy PDFs push token counts up quickly.
The key trade-off to watch is the gap between input and output pricing. Output tokens usually cost several times more per million, so a workload that only extracts short fields stays cheap, while one that rewrites or summarizes full documents can be dominated by output cost. Before committing, estimate tokens per page from a real sample rather than guessing, and check whether trimming the response length or processing fewer pages per document lowers the monthly total more than switching models does.
Cost per PDF = pages Γ tokens/page Γ input price + output tokens Γ output price. Feeding whole documents is input-heavy, so page count and how much you send drive the bill more than the answer length.
Don't send whole documents blindly β extract or retrieve only the relevant pages, chunk long PDFs, and cache repeated system prompts. Full-document input is the biggest, most avoidable cost.