Fulfillment Terms

Last updated February 22, 2024

This page provides the fulfillment terms for OpenPipe, Inc. (“Company,” “we,” “us,” or “our“) and describes how we charge for training and inference.

Training

OpenPipe charges for model training based on the architecture of the base model trained as well as the number of tokens the model will see during training. We provide an estimate of training costs at the time you start a model training job. This estimate is usually exactly right, but because of differences in the way tokenization is performed while preparing that estimate vs how it is done (in a more exact way) as part of the actual training job, the actual costs may differ.

We will never charge more than 2x the estimated costs when calculating the final cost of a training run.

Inference

Model inference performed through the OpenPipe API will be charged at a per-token rate as shown on our pricing page. Pricing varies by model.

Credits

Accounts may be issued credits for promotional or service-related reasons. Credits will generally be applied to the next invoices generated after the credit is generated. It is also possible to refund an existing invoice in full or in part if an error was made while calculating the invoice amount.

Billing

Billing is performed on a monthly cycle. At the end of each calendar month, each project will be issued a bill for the inference and training expenses accrued during that month. The bill will automatically be charged to the credit card on file, or if no credit card is on file the project owner will be emailed the invoice. In all cases an email invoice will be generated and sent to the project owner for their records.

Denial of Service for Payment Reasons

OpenPipe reserves the right to deny service to any project that has a balance below 0 if no valid credit card is on file. OpenPipe can also deny service for payment reasons if an account's usage is substantially higher than in previous months and there is no explicit conversation explaining the increased usage.

If you have any questions relating to payment or fulfillment of services, please contact us at support@openpipe.ai.

About OpenPipe

OpenPipe is the easiest way to train and deploy your own fine-tuned models. It only takes a few minutes to get started and can save you 25x relative to OpenAI with higher quality.

Introducing Direct Preference Optimization (DPO) Support on OpenPipe