Model Pricing

RouteAPI bills pay-as-you-go. Each model has its own price, and different kinds of models are billed in different ways: text models by tokens, some image and video models per image, per request, or per second. This page explains where to find prices and how each billing method works.

Where to find prices

Location	What you see
Model marketplace	All available models with input, output, and cache read prices, or per-request and per-second prices. Supports search and filtering by vendor and type.
Model detail page	Click a model in the marketplace to see its full price breakdown: cache write, image input, audio, resolution tiers, context tiers, and billing notes.
Usage log	The actual amount charged for each request, with tokens and request details.

• Prices are shown in US dollars (USD). Token prices use 1M tokens as the unit.
• A struck-through price next to a model is the official list price; the badge shows how much you save.
• If you are signed in and your account has an exclusive rate, the marketplace and detail pages show the price that applies to you.
• The displayed model rate is the full price. There is no additional platform fee.

Tip

Model availability and prices can change. Check the marketplace before a large-scale rollout, and use the usage log to confirm actual charges.

Text models

Text models (chat, reasoning, code, embeddings, and similar) are billed by tokens. The cost of one request is:

cost = uncached input tokens × input price
     + cached input tokens × cache read price
     + cache write tokens × cache write price
     + output tokens × output price

• Input / output: listed separately. Output is usually more expensive than input. Reasoning tokens from reasoning models count as output.
• Cache read: input tokens that hit the upstream prompt cache are billed at the cache read price, which is usually much lower than the regular input price. The part that misses the cache is billed at the regular input price.
• Cache write: some models (such as Claude) charge for writing to the cache. Writes with a longer retention (for example 1 hour) are priced higher and are listed separately on the detail page.
• Image and audio input: models that accept images or audio may list a separate image input or audio input/output price. Those parts are billed at their own rates.
• Embeddings: only input tokens are billed.

Context-tiered pricing

Some models price long-context requests higher. Such models are marked Tiered pricing in the marketplace, and the detail page shows a table such as ≤ 200K / > 200K with input, output, and cache prices for each tier.

The rules:

• The tier is chosen by the input-side tokens of a single request (including cached tokens). When they exceed a tier’s threshold, that tier applies.
• Once a higher tier applies, the whole request (input, output, and cache) is billed at that tier’s prices, not just the part above the threshold.
• Tiers are decided per request. They have nothing to do with your cumulative or monthly usage.

Note

RouteAPI does not offer automatic volume-based discounts. For enterprise pricing, contact sales.

Image models

Image generation models use one of the following methods. The detail page shows which one applies.

Billed by tokens

Models such as gpt-image and Gemini image models convert generated images into output tokens and bill them at the output price. The number of tokens per image depends on the output size and quality: larger and higher-quality images consume more tokens. Text prompts and reference images are billed as input tokens.

Billed per image

Other image models charge a fixed price per image. If the detail page shows a Size tiers table, the price depends on the requested size:

• The tier is chosen by the pixel count (width × height) of the requested size. 1K = 1024 × 1024, 2K = 2048 × 2048, and so on. Other aspect ratios are counted by actual pixels.
• Generating multiple images (n > 1) is billed per image.
• If size is omitted or cannot be recognized, the highest unit price applies. Pass size explicitly to control cost.
• Some models also charge for reference images beyond a free allowance. When this applies, the detail page lists it.

Video models

Video generation models use one of three methods:

Method	How it is billed	Shown as
Per request	Each generation has a fixed price, regardless of duration or resolution	$x / request
Per second	Billed by the seconds of generated video. Some models price higher resolutions or sizes higher	$x / second, with a resolution or size table
By tokens	Billed per 1M total tokens of the request. The unit price depends on the output resolution, whether the request includes video input, and whether audio is generated	A resolution × input type table, with an approximate ≈ $x / second reference under each price

For token-billed video, the per-second reference is only an estimate. The actual charge is based on the tokens reported by the upstream. Resolutions not listed in the table are billed at the price stated in the detail page notes.

Video generation is an asynchronous task:

1. When you submit the task, an estimated amount is reserved from your balance.
2. When the task completes, the charge is settled based on actual usage and any difference is returned.
3. If the task fails or times out, the reserved amount is refunded in full.

Audio, music, and tool calls

• Speech and realtime audio models: billed by tokens. Audio input and audio output have their own prices, shown separately on the detail page.
• Music (Suno) and Midjourney: billed per action at a fixed price, such as generating a song or lyrics, or each Midjourney operation.
• Built-in tools: when a request uses built-in tools such as web search, file search, or image generation, each tool call is billed separately on top of the token cost. These charges are included in the request’s amount in the usage log.

For billing fields and failure cases, see Usage and Billing.