Core formula
Monthly cost is the monthly interaction count multiplied by the estimated cost of one interaction. One interaction means one AI reply, one customer inquiry, or one character-app turn, depending on the selected scenario.
monthly cost = interactions / month × (LLM input + LLM output + cached input + TTS + vision)
Token and history estimates
Input and output token counts are user-supplied planning values, not text measurements. Retained conversation history is added to each new request. For Claude 4.7 and later, token counts can be roughly 30% higher than older tokenizers for the same text, so production measurements should replace the defaults.
Cached input
The cached share is moved from normal input billing to the provider's cache-read price when one is published. Cache writes, storage duration, and storage fees are excluded. If a default price record has no cache-read rate, cached text remains at the normal input rate.
Voice and Gemini TTS
Character-priced TTS is calculated directly from voiced characters. Token-priced Gemini TTS uses explicit conversion assumptions: 4 characters per text token, 15 characters per second of speech, and 32 audio tokens per second. These are planning assumptions, not provider guarantees.
Vision
Vision calls are represented as extra LLM input tokens. The default is 1,000 input tokens per vision call and can be changed in the livestream scenario.
What is excluded
Taxes, free tiers, subscription credits, volume discounts, rate limits, cache writes and storage, one-time voice fees, and provider-specific rounding are outside the estimate. Prices are stored with their source, status, and verification date.