We’ve introduced Private Link, providing customers with a secure means of direct communication between Azure and OpenAI while minimizing exposure to the open internet. Additionally, we’ve launched native Multi-Factor Authentication (MFA) to support compliance with evolving access control standards. These additions complement GPT existing suite of enterprise security features, including SOC 2 Type II certification, single sign-on (SSO), AES-256 data encryption at rest, TLS 1.2 encryption in transit, and role-based access controls. Moreover, we offer Business Associate Agreements for healthcare firms requiring HIPAA compliance, as well as a zero data retention policy for API clients with qualifying use cases.
Enhanced administrative control functionalities.
Through GPT latest Projects feature, organizations gain finer control and oversight over individual projects within OpenAI. This encompasses the capability to tailor roles and API keys to specific projects, regulate the availability of models, and establish usage and rate-based limits to manage access and prevent unforeseen overages. Additionally, project owners can generate service account API keys, providing project access without being linked to a particular user.
Enhancements to the Assistants API
We’ve implemented several updates to the Assistants API to enhance accuracy in information retrieval, provide more flexibility in model behavior and task completion tools, and offer better cost control. These updates include:
- Improved Retrieval with ‘file_search’: Introducing ‘file_search’ capability allowing ingestion of up to 10,000 files per assistant—a significant increase from the previous limit of 20 files. This tool offers faster performance, supports parallel queries with multi-threaded searches, and features enhanced reranking and query rewriting.
- Streaming Support for Real-time Responses: Fulfilling a top request from developers and enterprises, we’ve added streaming support for real-time, conversational responses.
- ‘vector_store’ Objects: Introducing ‘vector_store’ objects in the API, enabling files to be added to a vector store and automatically parsed, chunked, and embedded for file search preparation. These vector stores can be utilized across assistants and threads, simplifying file management and billing.
- Control Over Token Usage: Providing control over the maximum number of tokens used per run, along with limits on previous and recent messages utilized in each run, to help manage token usage costs effectively.
- ‘tool_choice’ Parameter: Introducing the ‘tool_choice’ parameter, allowing selection of a specific tool (such as ‘file_search’, ‘code_interpreter’, or ‘function’) in a particular run.
- Support for Fine-tuned GPT-3.5 Turbo Models: Adding support for fine-tuned GPT-3.5 Turbo models in the API. Initially, we’ll support fine-tunes of ‘gpt-3.5-turbo-0125’.
Expanded options for managing costs
To support organizations in efficiently scaling their AI usage while managing costs effectively, we’ve introduced two new features:
- Discounted Usage on Committed Throughput: Customers maintaining a consistent level of tokens per minute (TPM) usage on GPT-4 or GPT-4 Turbo can now request access to provisioned throughput, unlocking discounts ranging from 10% to 50% based on the size of the commitment.
- Reduced Costs on Asynchronous Workloads: Introducing GPT new Batch API, allowing customers to execute non-urgent tasks asynchronously. Batch API requests are priced at a 50% discount compared to shared prices, offer higher rate limits, and deliver results within 24 hours. This feature is particularly beneficial for use cases such as model evaluation, offline classification, summarization, and synthetic data generation.
We are committed to continually enhancing GPT platform with additional features focused on enterprise-grade security, administrative controls, and cost management. For further details on these releases, please refer to GPT API docume
enterprise needs.