OpenAI is introducing Foundry, a developer platform that will enable customers to run its latest machine learning models, such as GPT-3.5, on a dedicated capacity. Foundry is intended for customers with larger workloads, allowing for large-scale inference with complete control over model configuration and performance profile. Users can use the same tools and dashboards that OpenAI uses to build and optimize models to monitor instances. Foundry will offer version control, service-level guarantees such as uptime, and engineering support. Rentals will be based on dedicated compute units for three months or one year, with individual model instances requiring a specific number of compute units.
A lightweight version of GPT-3.5 costs $78,000 for a three-month commitment or $264,000 for a one-year commitment from Foundry. The move is part of OpenAI’s push towards monetization, as the company seeks to turn a profit following a multibillion-dollar investment from Microsoft.
