Expanded launch of enterprise-grade features for API customers

OpenAI has introduced new features focused on security, administrative controls, and cost management for enterprise customers and rapidly growing developers.

For enhanced security, OpenAI launched Private Link, which supports direct communication between Azure and OpenAI, and MFA (multi-factor authentication). In addition to the existing SOC 2 Type II, SSO, data encryption (AES-256, TLS 1.2), and RBAC, it also offers BAA for the healthcare sector and a zero data retention policy for specific use cases.

The Projects feature allows organizations more granular control over individual projects. Roles and API key scopes can be restricted per project, available models can be specified, and usage and rate limits can be set. It also enables the creation of service account API keys that are not tied to individual users.

The Assistants API has received significant improvements as follows:

• file_search: Supports up to 10,000 files (a 500x increase from the previous 20), with parallel querying and improved reranking capabilities.
• Streaming and vector_store: Supports streaming for real-time conversational responses, and introduces the vector_store object, which automatically parses and embeds files.
• Cost and tool control: Added a maximum token usage limit per run and a tool_choice parameter that allows selecting specific tools.
• Support for GPT-3.5 Turbo fine-tuned models begins.

Two options have been added to improve cost efficiency: a Committed throughput discount (10-50%) for customers with sustained usage, and the Batch API, which allows non-urgent tasks to be processed at a 50% lower price.