How many model calls per month.
0% of input tokens reused (system prompts, RAG context). Major savings on OpenAI / Anthropic / Gemini.
~750 tokens ≈ 1 page of text.
Output is typically 3-6× more expensive than input.
Cheapest$18.00/moGPT-4o mini
Most expensive$2,100/moClaude Opus 4
Spread116.7×cheapest → priciest
Prices reflect public list prices and change frequently. Volume discounts, batch APIs (50% off on OpenAI/Anthropic), and prompt caching can drop your real cost meaningfully.