AI Model API Cost Calculator

Your usage

Reference: estimated monthly cost at default usage

Assumptions: 100 calls/day · 2,000 input tokens per call · 800 output tokens per call · off-peak rates on · 30% cache hit rate · 30-day month · amounts in the original billing currency (CNY). Results are estimates, not purchase advice.

Models without a published cache price are estimated at the standard input price (no cache discount assumed); time-based priced models use the off-peak rate (matching the "Off-peak" toggle above); tiered models are priced at the tier an example input of 2,000 tokens falls into. The estimate column uses exactly the same formula, default parameters and reference rate as the interactive calculator below. USD is the official billing currency for those models; estimated monthly costs are converted to CNY (1 USD = 6.718 CNY, CFETS central parity rate 2026-10-23), indicative only. For a two-model, line-by-line comparison use the model comparison.

Variables, formula and a worked example

Not covered: free tiers, resource packs, subscription discounts and limited-time promos are not included; tiered models are estimated at the tier your input length falls into; an estimate is not a vendor's actual invoice and does not constitute purchase advice.

Frequently asked questions

How does this calculator work?

Monthly cost = calls per day × 30 × (input tokens ÷ 1,000,000 × effective input price + output tokens ÷ 1,000,000 × output price). The effective input price is weighted by cache hit rate: hit rate × cache-hit price + (1 − hit rate) × standard price. Models without a published cache price are billed at the standard input price.

When do DeepSeek off-peak rates apply?

DeepSeek uses time-based pricing. Peak hours are weekdays 9:00–12:00 and 14:00–18:00; all other times (nights, early mornings, weekends and public holidays) use the off-peak rate, which is about half the peak rate. Scheduling batch jobs off-peak directly reduces cost.

What does cache hit rate mean?

The cache hit rate is the share of input tokens that hit the context cache. Hit tokens are billed at a much lower cache-hit price; the rest are billed at the standard input price. Keeping a stable prompt prefix (such as a fixed system prompt) raises the hit rate and lowers input cost.