Pricing

HyperFlow is the seller on the marketplace today and charges each model at its list price, with no subscription. Current prices are on the Models page and each model’s own page.

Units

A model’s type decides the unit it’s priced in.

Tiers

Some models have more than one price:

• Chat models can be priced by context length, so long prompts cost more per token. The Models page shows the lowest price as “from $X”; the model’s page lists each tier.
• Cached input: models that support prompt caching list separate prices for cache reads and writes.
• Video models can be priced by resolution.

What a request costs

A request costs its usage times the price: tokens in and out for chat and embeddings, images for image models, seconds for video. The Logs tab of Usage shows each request’s usage and the exact price applied. Prices are in credits, where 1 credit = $1 of usage.

For example, at an input price of 0.1 and an output price of 0.5 per 1M tokens, a chat request that reads 2,000 tokens and writes 500 costs 2,000 × 0.1 / 1M + 500 × 0.5 / 1M = 0.00045 credits.

The hold

Before a chat request runs, HyperFlowSI holds the most it could cost from your balance: its prompt plus the most output it could produce. That output is your max_tokens for each of the n choices, or a default allowance when you don’t send max_tokens, and never more than the model’s own output limit. The same limit is passed to the model, so an answer can’t run past what was held.

• Set max_tokens to control the hold, and to give a long answer room to finish.
• When the request finishes, you’re charged for what it used and the rest is released.
• If your available balance can’t cover the hold, the request is refused with 402.

Image and video requests are held for the images or seconds you ask for.