What are AI tokens and how are they billed?

AI models do not read words. They convert text into numbers called tokens. Some long words count as two tokens and punctuation counts too. As a rule of thumb, about 750 words equal 1,000 tokens, and a 12 word greeting with punctuation is about 13 tokens.

For AI agents, token costs are billed as pass-through: what we pay the AI provider (OpenAI, Google or whichever model you choose) is what you pay, with no markup or margin. On the default fast model this is about eight one-hundredths of a cent per 1,000 tokens. More powerful models cost more, around a penny per 1,000 tokens for the most powerful ones, which is why they suit non-voice work better.

A voice agent is charged per minute of conversation plus its tokens. Because tokens cost only a fraction of a penny per 1,000, the token part is very small, so in practice a voice agent costs the per-minute price plus a little. The pricing calculator on the pricing page at votel.ai shows the cost for the model you pick.