Pay-as-you-go is how most cloud services, and most LLM APIs, charge: a price per unit and a meter.
How it works
AWS describes the model as paying “only for the individual services you need, for as long as you use them, and without requiring long-term contracts.” LLM APIs apply it per token, with separate rates per 1M tokens for:
- new input tokens;
- cached input tokens, when the provider offers caching;
- output tokens, including reasoning.
Speech-to-text is usually priced per audio minute instead.
A monthly bill is the sum over requests:
input × input rate + cached × cached rate + output × output rate
Billing usually runs from a prepaid balance that requests draw down, or from a monthly invoice. Each request’s usage block tells you exactly what was metered.
Why it matters
- Low entry cost. Nothing to pay while idle, so it suits prototypes, light apps and bursty traffic.
- Cost tracks value. Each request has a known price, which helps when you charge your own customers per use.
- Less predictable. An agent stuck in a retry loop, a long context resent on every turn or a burst of reasoning tokens can multiply a bill. Heavy steady use can cost more than a flat-rate plan.
On BroadNet Inference
New accounts start with a free $5 credit. BroadNet Gemma 4 VL 26B and both transcription models work on it; DeepSeek V4.1 Flash, GLM 5.3 and Qwen3.8 27B need a top-up. Sample rates per 1M tokens:
| Model | Input | Output |
|---|---|---|
| DeepSeek V4.1 Flash | $0.175 | $0.52 |
| BroadNet Gemma 4 VL 26B | $0.10 | $0.22 |
Arabic speech to text costs $0.0042 per audio minute. A pay-as-you-go key allows 30 requests a minute and 4 in flight. There is also an hourly key: $0.50 for one hour of BroadNet Gemma 4 VL 26B on one stream.
All rates are on the pricing page, with a calculator. In BroadNet’s worked example, a side project with 100 million input tokens a month costs $14.23 on DeepSeek V4.1 Flash, so pay as you go wins. A full-time coding agent costs $65.40, so Full Moon wins. The math is in flat rate vs pay per token.