---
title: "Hedra pricing"
description: "Custom Model Factory engagement pricing and GPU compute rates."
canonical: "https://www.hedra.com/pricing"
---

# Hedra pricing

[View on Hedra](https://www.hedra.com/pricing)

## Model Factory

**Custom pricing per engagement.**

Build or improve a model with our forward deployed researchers.

[Talk to our team](https://www.hedra.com/enterprise?form=true)

[View GPU rates](https://www.hedra.com/pricing#gpu-rates)

## GPU compute

| GPU | Rate |
| --- | --- |
| NVIDIA B300 ($8.25 / GPU-hour equivalent) | $0.00229167 per GPU-second |
| NVIDIA B200 ($7.26 / GPU-hour equivalent) | $0.00201667 per GPU-second |
| NVIDIA RTX PRO 6000 ($2.49 / GPU-hour equivalent) | $0.00069167 per GPU-second |
| NVIDIA H100 ($3.00 / GPU-hour equivalent) | $0.00083333 per GPU-second |
| NVIDIA RTX 4090 ($0.89 / GPU-hour equivalent) | $0.00024722 per GPU-second |
| AMD Instinct MI355X ($5 / GPU-hour equivalent) | $0.00138889 per GPU-second |
| AMD Instinct MI455X (Research preview) | Contact us for pricing |

Your quote confirms supported hardware, rates, included resources and billing terms. Research and negotiated reservations are scoped separately.

[Discuss your workload](https://www.hedra.com/enterprise?form=true)

[Model API rates](https://www.hedra.com/develop/models)

## Billing FAQ

### How are production inference, model training and GPU Sandboxes priced?

GPU compute is priced per GPU-second. Your quote confirms the supported hardware, rates, included resources and billing terms. Model training and GPU Sandboxes are research previews.

### Is dedicated inference the same as reserved GPU capacity?

Dedicated inference describes how your model is served. A reservation is a commercial commitment for agreed GPU capacity and a term. Confirm the supported workload, allocation, start date and billing terms with our team.

### How is a Model Factory engagement priced?

Model Factory combines a scoped research engagement with compute for experiments and production serving. Your proposal identifies the research scope and compute charges. Bring us the problem you need to solve; you don’t need a finished model or complete dataset to start.

### Can I buy compute without a research engagement?

Yes. You can discuss production inference or compute directly, or scope research and model development through Model Factory. Model training and GPU Sandboxes remain research previews.

### Where can existing Model API customers find rates?

View Model API rates for current model-specific prices and configurations. Existing Model API calls use a prepaid USD wallet, separate from Play subscriptions. These catalog prices do not apply to custom-model compute or research engagements.

### Do Play subscriptions and API usage share credits?

No. Play subscription credits and the API wallet are separate. The REST API and CLI use the API wallet. MCP uses the billing mode selected in your MCP settings.

[Play pricing](https://www.hedra.com/play/pricing)