Estimate cost and measure latency

Pricing is a changeable snapshot

On September 19, 2026, TypeSafe's model page lists $0.042 per million input tokens and no output charge. The calculator uses this as an editable starting point. Provider rates and account conditions can change; confirm them in your actual account.

Understand the formula

Estimated cost = calls × billed input tokens per call ÷ 1,000,000 × input rate.

For example, 10,000 calls at 2,500 input tokens each total 25 million tokens, or about $1.05 at the example rate. This is arithmetic under chosen assumptions, not a recorded bill.

Count billable input using your provider's rules, not just the customer's sentence. State, questions, option descriptions, retries and repeated requests can affect usage. A batch of questions is not automatically multiple full requests; shared state is not automatically free either.

Exclusions

OCR, transcription, generative models, search services, storage, infrastructure, transfer, taxes and human review are not included in the model input price. Check each provider's billing terms separately.

Measure latency at two levels

Record request-to-response time and the full user-input-to-usable-result duration. Retrieval, networking, queuing, preprocessing, retries and actions affect the experience.

Test sequential calls first, then realistic concurrency and batching. Report sample count, region, provider, model version and p50 / p95. A single fast response is not a universal performance guarantee.