OpenAI Foundry lets customers purchase dedicated computing to run AI models

OpenAI quietly launches a new platform for developers that allows customers to run newer machine-learning models from the company, such as GPT-3.5, on dedicated capacity. OpenAI’s Foundry is described as \”designed for cutting edge customers running large workloads\” in screenshots posted on Twitter by early access users.

The documentation states that \”[Foundry] allows] inference on scale with full control of the model configuration and profile,\” OpenAI has been contacted to confirm this information.

According to the screenshots, Foundry will, when it launches, provide a \”static allocation\” (possibly on Azure, OpenAI’s preferred public cloud platform), dedicated to one customer. The same dashboards and tools that OpenAI use to optimize and build models will be available to users to monitor individual instances. Foundry also offers a level of versioning, which allows customers to decide whether to upgrade to the latest model releases. It will also provide \”more robust\” fine tuning for OpenAI’s newest models.

Source:

OpenAI’s Foundry will let customers buy dedicated compute to run its AI models