Skip to content

Toolsabout 1 min

What will an AI assistant cost to run each month?

Real per-token prices from Anthropic, OpenAI and Google, applied to your conversations: the monthly bill, the cost of one conversation, and what changes it most.

Change the highlighted values

Every month the assistant has conversations of about messages each. The messages are , in , and the answers draw on . Prompt caching is .

Typical volumes

Compare models at your volume

Anthropic
OpenAI
Google

Prices checked: 7 October 2026 · Sources: Anthropic · OpenAI · Google

The full report by email

  • The full breakdown behind the figure
  • What we would do first
  • A link to book a call about it
Privacy policy

How we count

  1. 1Each message the assistant answers sends the model its instructions (1 500 tokens), any document excerpts (none, 3 000 or 12 000 tokens), the conversation so far and the new message; the model then writes its reply. Long conversations cost more per message, because the history is sent again every time.
  2. 2Words become tokens at about 2 tokens a word in Lithuanian and 1,35 in English (our measurement, rounded). Anthropic states its newer models (Claude 4.7 and later) use about 30 % more tokens for the same text, so we add that for Sonnet 5.5 and Opus 5.5.
  3. 3With prompt caching, the instructions and the history are billed at the cached-input price; each new exchange is written to the cache (Anthropic and OpenAI charge 1,25 × the input price for that; Google does not). Document excerpts change with each question and are billed in full.
  4. 4Prices are each provider's public list price per million tokens, standard tier, checked on 7 October 2026, converted at the ECB reference rate of 1 EUR = 1,1269 USD. Batch processing (50 % off) is not counted: it does not fit live conversations.
  5. 5Not included: hosting, monitoring and your own staff's time. Gemini 3.8 Flash is at a promotional price until the end of 2026; Gemini 3.1 Pro is still a preview model.

Sources

Want to put this in place?

A free 30-minute call: we pick the model and the set-up for your assistant, with its running cost measured on your own questions.

Book a call