AI API monthly cost calculator

Estimate AI inference and token costs from your request volume, input/output tokens, caching and batch usage. Compare up to four models and their channels in USD. This is a budget calculator, not a connected billing dashboard.

1. Usage

Usage period

Daily usage is converted using a 30-day month.

Requests / day

Avg input tokens

Avg output tokens

Cache hit rate0%

Channels without a published cache price use the regular input price.

Batch request share0%

Assumes the common 50% Batch discount; confirm channel eligibility before use.

30K

requests/mo

30M

input/mo

15M

output/mo

2. Models and region

Qwen3 Coder Flash1/4

Monthly cost comparison

Paid channels are sorted by USD-normalised monthly cost.

Model	Channel	Input price	Output price	Cache	Est. monthly
Qwen3 Coder Flash	Qwen(阿里)China	¥1.00/1M	¥4.00/1M	—	$13.04Lowest paid¥90.00 native
Qwen3 Coder Flash	OpenRouterGlobal	$0.20/1M	$0.98/1M	$0.04	$20.48USD normalised

Codex subscription weekly reference (separate from API cost calculations)

Astra / Sol post-upgrade retest · ChatGPT Pro 200

Quota verified on: 2026-10-30

Model / successful calls	Measured tokens	Weekly capacity (conditional)
gpt-6-astra96/96 successful	8,922,563	111.6M–186.0M tokens/week; point estimate 139.5M
gpt-6.1-sol94/96 successful	8,729,647	≥ 274M tokens/week (conditional lower bound)

The upgrade is user-reported and has not been independently verified.

Applies only to ChatGPT Pro 200, not Pro 100/500.

Only the weekly window is available; no extrapolation to 5-hour or monthly windows.

The two models share one allowance pool; capacities cannot be added.

The historical and current Astra intervals overlap; interpret results with their workload mix.

The Sol lower bound assumes settled linear metering, nonnegative extra consumption and no refunds, resets or allowance changes; the main batch included controller activity.

10-30 retest report (Chinese)10-30 meter readings and per-call ledger 10-28 historical data and report (Chinese)