BILLING
Token pricing
How the cost of a request follows from its tokens and the published price.
Tessen's model families are billed per token from a prepaid wallet, in US dollars. Prices are per one million tokens (price version commercial-v1-2026-10-25) and are published without a key at GET /v1/pricing/families.
| Model | Model id | Input | Cached input | Output |
|---|---|---|---|---|
| Z1 | z1 | $1.50 | $0.15 | $6.00 |
| Z1 Code | z1-code | $2.00 | $0.20 | $5.00 |
| H1 | h1 | $8.00 | $0.80 | $30.00 |
The formula
TEXT
cost = (uncached_input_tokens × input_price + cached_input_tokens × cached_price + output_tokens × output_price) ÷ 1,000,000- Input tokens are
usage.prompt_tokens: your messages, system prompt, tool definitions and tool results. - Output tokens are
usage.completion_tokens. Reasoning tokens are output tokens, billed at the output price. - Cached input tokens, when the model reports them, are billed at the cached price.
Worked examples
| Request | Tokens | Calculation | Cost |
|---|---|---|---|
| Z1: "Reply with the single word: ok" | 11 in, 2 out | 11 × $1.50 ÷ 1M + 2 × $6.00 ÷ 1M | $0.0000285 |
| H1: the same message | 19 in, 176 out (mostly reasoning) | 19 × $8.00 ÷ 1M + 176 × $30.00 ÷ 1M | $0.005432 |
Both costs are the ones the Gateway reported in the final streamed frame (cost_usd) and debited from the wallet.
What costs nothing
- Project ZERO Alpha requests: free within the daily allowance; the wallet is never debited.
- Embeddings: not metered.
- Refused requests: a request that is refused (401, 402, 403, 413, 422, 503) is not charged.
Note
The Console shows costs smaller than a cent with up to seven decimals, so a tiny charge is never displayed as $0.00.