How it works

1

Monthly inferences

Estimated active users x 30 days x trigger frequency per day. This is the total number of times your model is called in a month.

2

Cost per inference

Model call cost varies by feature type. Text summary uses a language model (~$0.002/call). Document scan uses a vision model (~$0.005/call). Recommendation uses a smaller model (~$0.0002/call). Smart search requires embedding plus generation (~$0.003/call). Voice uses speech-to-text and TTS (~$0.004/call).

3

Break-even price

The minimum you need to charge per paying user to cover the inference cost. Assumes a 30% conversion rate (industry median for freemium mobile apps), so 30% of your active users are paying users.

Questions

What formula does this use?

Monthly bill = active users x 30 x trigger frequency x cost per inference. Cost per user = monthly bill divided by active users. Break-even price = monthly bill divided by active users x 0.3, where 0.3 is the 30% conversion rate assumption.

How accurate are the cost figures?

They are 2024 list prices for leading cloud AI providers. Actual costs depend on prompt length, response length, and model version. Use these for order-of-magnitude planning, not procurement.

What is the break-even price?

The minimum monthly charge per paying user to cover the AI inference cost. It assumes 30% of your active users are paying users. If your conversion rate is higher, your break-even price is lower.

Can I share or save my results?

Yes. The Share link button encodes your inputs in the URL. Anyone with the link sees the same result.

Should I use on-device or cloud AI?

For recommendation and ranking, on-device can cut cost by 80-95% with acceptable accuracy. For generation tasks (summarise, voice), cloud is currently required unless you can use a small distilled model.

Who should I talk to if my situation is more complex?

Book 30 minutes with one of our engineers. You will leave with a cost model for your specific feature, model choice, and usage pattern.

Related calculators

Related reading

Have a complex scenario? Talk to an engineer →