You pay for what our decision models read.
You pay for input tokens, monthly, with no minimum or commitment.
$0.03
per million input tokens
Output: freeEvery answer comes back as probabilities, not generated text.
- Model
decide-1.1, through one HTTPS endpoint. Every answer carries a calibrated probability for each option.- What counts
- The document, once per request, plus the words of each question. Asking three questions about a document doesn’t triple what you pay.
- Billing
- Monthly, for the month’s usage, by card through Stripe. The console shows the month so far, updated every 15 minutes.
- Commitment
- None. Revoke a key or end billing at any time.
- Currency
- US dollars, before any taxes that apply.
- Your own models
- Train a Fast, Edge or Smart model on your team’s past decisions in the console: $0.40 per million training tokens on Fast and Edge, $1.00 on Smart, with a $2 minimum a run. You confirm an exact quote before a run starts and never pay more than it. The held-out evaluation is free. Calling a Fast model is free; an Edge model costs $0.02 and a Smart model $0.03 per million input tokens, the same as decide-1.1.
What a workload costs
| Work | Documents | Questions each | Input tokens | Cost |
|---|---|---|---|---|
| Support tickets | 1,000,000 of 500 tokens | 1 | 500,000,000 | $15 |
| Insurance claims | 1,000,000 of 1,000 tokens | 3 | 1,000,000,000 | $30 |
| Contracts | 100,000 of 8,000 tokens | 2 | 800,000,000 | $24 |
The figures count the documents. Each question adds its own words, usually a few dozen tokens.
This price has applied since .