Pay-as-you-go LLM pricing, explained

What pay-as-you-go actually means

Pay-as-you-go (PAYG) LLM pricing means you pay for the tokens you use, nothing else. No monthly seat fee, no prepaid tier you have to grow into, no credit that expires. On Riven, standard lanes are a flat $3.00 per 1M tokens, and you can start at $0 with no minimum.

This is distinct from the two other common models in the market: per-seat subscriptions, where you pay a fixed monthly amount per user regardless of usage, and prepaid token credits, where you buy a block of tokens upfront and draw it down. PAYG is the third option — you are billed for what you consume, after you consume it, at a rate you can see in advance.

The numbers

The Riven pricing model is deliberately simple:

• Standard lanes: $3.00 per 1M tokens flat. This covers the bulk of the 160-model catalog.
• Frontier-class models: vendor list rates, transparent per-token pricing. For example, Kimi K3 bills at Moonshot's published $3 in / $15 out per 1M tokens — the same rate you would pay the vendor directly.
• $0 to start, no minimum. No deposit, no committed spend to get an API key.
• No seat minimums. One developer or a hundred, the rate is the same per token.

The full datasheet is at /pricing and /developers/pricing, and the numbers there match your invoice exactly.

How metering works

Metering is per-token, and it is visible. Every API call reports token usage in the standard OpenAI response shape, and that usage is what you are billed on:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.rivenai.io/v1",
    api_key="rk_live_...",
)

resp = client.chat.completions.create(
    model="qwen-3.5-397b",
    messages=[{"role": "user", "content": "Explain backpressure in two sentences."}],
)

# usage is reported per-token; this is what you are billed on
print(resp.usage.prompt_tokens, resp.usage.completion_tokens)

Because the usage object is the source of truth for billing, you can compute your expected cost from your own logs without trusting a dashboard. The rate from the datasheet times the tokens from the response equals the charge. There is no second "real" rate sheet.

When PAYG beats a subscription

PAYG is the right model when usage is variable, hard to predict, or spiky:

• Variable workloads. A batch job that runs once a week, a feature with seasonal traffic, an internal tool used inconsistently. A subscription charges you the same in a quiet month as a busy one; PAYG charges you for what you run.
• Prototyping and evaluation. When you are comparing models or building something new, you do not know your usage yet. PAYG means you do not commit to a tier you might outgrow or underuse.
• Cost attribution. Per-token billing maps cleanly to per-feature or per-customer cost. You can attribute spend to the call that caused it, which is hard to do honestly with a flat seat subscription.

A subscription wins when usage is high and predictable enough that a committed tier gives a meaningful discount. For everything else — which is most workloads, especially early — PAYG is the cheaper and more honest option.

What you give up

Nothing, structurally. PAYG on Riven is the same API, the same 160+ models, the same single-tenant enterprise path governed by Riven Control. The pricing model is independent of the capability. You do not get a lesser product by paying per token; you get the same product billed differently.

Getting started

The path is short: sign up, get a key, point your OpenAI SDK at https://api.rivenai.io/v1. Standard lanes bill at $3.00 per 1M tokens from the first call. If you want to try models before integrating, the web chat at rivenai.io gives 10 free guest prompts with no account.

Get started

Create an account at /sign-up to get an API key with $0 minimum, or read the docs for the full pricing and metering reference.