Pricing

Introduction

You pay for what each request uses, in USD, with no subscription and no minimum spend. Your balance is deducted as you generate and it never expires.

Two things decide what a generation costs: the model you call and what you ask it for. A 10 second video costs more than a 5 second one, and a 2048 × 2048 image costs more than a 512 × 512 one.

How a model is priced

Most models charge a rate per unit. The unit is whatever the model produces or consumes, from a second of video to a million tokens. Multiply the units you asked for by the rate and you know the price before you send the request.

Some models are billed by compute time. The charge follows the GPU time your request uses, so every setting that gives it more work to do makes it cost more. These models publish measured runs instead of a rate: real generations, each with the settings that produced it and what it was charged.

Compute time also works in your favor. When an optimization cuts the GPU time a model needs, the cost of a generation drops with it, with no change on your side and no new price to adopt.

What changes the cost

Anything that gives the model more work to do costs more. These are the usual ones, and they are examples rather than the whole list.

A model can charge for other things too, such as each reference image past the first few, or the characters of a script it has to read. The rate card on the model's page names every unit that model bills by.

Where the price of a model lives

A model with a published price carries a Cost and speed panel on its page, with the rates, the measured runs, and example generations with their price and the time they took. When a model is running a temporary offer, a badge shows it there, and what you are charged while the offer lasts is lower than the listed rate.

Start at the Models section. A model with a rate tells you what a request costs before you send it. One billed by compute time tells you after it runs, and its measured runs are there to size it up first.

Pricing data as JSON

The same data is published as JSON, so an agent or a script can read prices without scraping a page.

Ask any model for its pricing block and you get the shape that fits how it charges. The example lists are shortened here.

curl -s https://runware.ai/docs/models/alibaba-wan3-0/schema.json | jq '.info["x-pricing"]'

{
  "currency": "USD",
  "docs": "https://runware.ai/docs/platform/pricing",
  "rates": [
    { "amount": 0.05, "unit": "durationSecond", "label": "480p", "display": "$0.05 per second" },
    { "amount": 0.1, "unit": "durationSecond", "label": "720p", "display": "$0.10 per second" },
    { "amount": 0.2, "unit": "durationSecond", "label": "1080p", "display": "$0.20 per second" }
  ],
  "examples": [
    {
      "id": "mycelium-packaging-factory-b-roll",
      "configuration": "first-frame · 720p · 5s",
      "price": 0.504,
      "latencyMs": 127753
    }
  ]
}

curl -s https://runware.ai/docs/models/prunaai-p-image-ideogram/schema.json | jq '.info["x-pricing"]'

{
  "currency": "USD",
  "docs": "https://runware.ai/docs/platform/pricing",
  "rates": [
    { "amount": 0.003, "unit": "output", "label": "1K - very low thinking", "display": "$0.003 per run" },
    { "amount": 0.006, "unit": "output", "label": "2K - very low thinking", "display": "$0.006 per run" }
  ],
  "examples": [
    {
      "id": "stage-backdrop-conservation-book-cover",
      "configuration": "text-to-image",
      "price": 0.0225,
      "latencyMs": 16427
    }
  ]
}

curl -s https://runware.ai/docs/models/minimax-m3/schema.json | jq '.info["x-pricing"]'

{
  "currency": "USD",
  "docs": "https://runware.ai/docs/platform/pricing",
  "basis": "token",
  "rates": [
    { "amount": 3e-07, "unit": "inputToken", "label": "≤ 512K context", "display": "$0.30 per 1M input tokens" },
    { "amount": 6e-07, "unit": "inputToken", "label": "> 512K context", "display": "$0.60 per 1M input tokens" },
    { "amount": 1.2e-06, "unit": "outputToken", "label": "≤ 512K context", "display": "$1.20 per 1M output tokens" },
    { "amount": 2.4e-06, "unit": "outputToken", "label": "> 512K context", "display": "$2.40 per 1M output tokens" },
    { "amount": 6e-08, "unit": "cachedInputToken", "label": "≤ 512K context", "display": "$0.06 per 1M cached tokens" },
    { "amount": 1.2e-07, "unit": "cachedInputToken", "label": "> 512K context", "display": "$0.12 per 1M cached tokens" }
  ],
  "examples": [
    {
      "id": "forced-utc-time-function-call",
      "configuration": "tool-calling",
      "price": 0.000136,
      "latencyMs": 1230
    }
  ]
}

curl -s https://runware.ai/docs/models/alibaba-z-image/schema.json | jq '.info["x-pricing"]'

{
  "currency": "USD",
  "docs": "https://runware.ai/docs/platform/pricing",
  "measured": [
    { "configuration": "512×512 · steps 20", "price": 0.0013, "latencyMs": 6175 },
    { "configuration": "1024×1024 · steps 20", "price": 0.0045, "latencyMs": 16846 },
    { "configuration": "1024×1024 · steps 50", "price": 0.0102, "latencyMs": 20374 }
  ],
  "examples": [
    {
      "id": "chicken-embryo-cutaway-workbook-illustration",
      "configuration": "inpaint · 896×1152",
      "price": 0.0045,
      "latencyMs": 19568
    }
  ]
}

A model billed by compute time has no rates. It carries measured instead, with the price of each configuration we ran. Most of both kinds carry examples as well, the runs behind those numbers.

The cost of a request you send

Add includeCost to any task and the response carries a cost field with the USD amount deducted for it.

import { createClient } from '@runware/sdk'

const client = await createClient({ apiKey: process.env.RUNWARE_API_KEY })
await client.connect()

const [result] = await client.run({
  model: 'runware:z-image@0',
  positivePrompt: 'a red apple on a wooden table',
  width: 1024,
  height: 1024,
  includeCost: true
})

import asyncio
import os

from runware import Runware


async def main():
    async with Runware(api_key=os.environ["RUNWARE_API_KEY"]) as client:
        results = await client.run({
            "model": "runware:z-image@0",
            "positivePrompt": "a red apple on a wooden table",
            "width": 1024,
            "height": 1024,
            "includeCost": True
        })


asyncio.run(main())

curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "imageInference",
      "taskUUID": "39d7207a-343e-4198-9c40-ef6d302cadb3",
      "model": "runware:z-image@0",
      "positivePrompt": "a red apple on a wooden table",
      "width": 1024,
      "height": 1024,
      "includeCost": true
    }
  ]'

runware run runware:z-image@0 \
  positivePrompt="a red apple on a wooden table" \
  width=1024 \
  height=1024 \
  includeCost=true

{
  "taskType": "imageInference",
  "taskUUID": "39d7207a-343e-4198-9c40-ef6d302cadb3",
  "model": "runware:z-image@0",
  "positivePrompt": "a red apple on a wooden table",
  "width": 1024,
  "height": 1024,
  "includeCost": true
}

Failed generations are not charged. You only pay for tasks that returned a result.

Balance and payments

All amounts are in USD. The Dashboard is where you top up, turn on auto-reload so your balance refills when it falls below a threshold, set low balance alerts, and add a backup payment method.

For high volume work, talk to sales about custom pricing.