OpenAI API, hosted on Microsoft Azure, provides users with access to the latest pre-trained AI models developed by OpenAI, including DALL-E, Codex, and GPT-4. It is designed to enable users to use its advanced AI capabilities to perform virtually any task described in the English language. However, because OpenAI is continually working on its API, this technology is constantly growing, and more advanced features are available.
In this article, we will talk you through some of those advanced features and capabilities. Many of them were introduced very recently, in April 2024. Let’s see what OpenAI has in store.
Improved security features
If you are an enterprise user, you will certainly be interested in the fact that OpenAI API now has new enterprise-grade security features, allowing the most demanding users to use this technology not just effectively but also safely.
The truth is that OpenAI API was quite secure even before April 2024. Up to April, OpenAI’s security technologies included:
- 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.
What’s especially interesting, OpenAI also offers Business Associate Agreements (BAAs) that are available for healthcare companies requiring HIPAA compliance.
Now, on top of all that, OpenAI introduces two more security features:
- Private Link: This feature ensures secure, direct communication between Azure and OpenAI, minimizing exposure to the open internet and enhancing data protection.
- Multi-Factor Authentication (MFA): Native MFA support helps organizations comply with increasing access control requirements, adding an extra layer of security.
Better administrative control
Here, we especially need to mention the new Projects feature. It was designed to provide users (again, especially large organizations) with more granular control and oversight over diverse projects that are in progress. With this new feature, users can:
- Assign specific roles and API keys to particular projects, ensuring precise access control.
- Restrict or allow access to specific models as needed.
- Set limits to manage access and avoid unexpected overages.
- Create API keys that provide access to projects without being tied to individual users, thus enhancing flexibility and security.
This is surely a positive change; with the Projects feature, you get better control over your work and enjoy greater flexibility.
Assistants API improvements
The Assistants API is a part of the OpenAI API suite that allows users to create, customize, and manage AI assistants (chatbots) tailored to specific tasks. In April 2024, OpenAI also introduced many important changes to this part of their ecosystem, making it more accurate and flexible.
Here are some of the latest improvements you should be aware of if you use this tool:
- Improved retrieval: The new file_search tool can process up to 10,000 files per assistant, which is a 500x increase from the previous limit. Moreover, this feature now supports parallel queries, multi-threaded searches, and features enhanced reranking and query rewriting.
- Streaming support: Real-time, conversational responses are now supported.
- Vector store objects: The new vector_store object allows files to be added to a vector store, automatically parsed, chunked, and embedded for efficient file search. This simplifies file management and billing across assistants and threads.
- Token usage control: Control the maximum number of tokens used per run and limit previous and recent messages in each run to manage costs effectively.
- Fine-tuned models: Support for fine-tuned GPT-3.5 Turbo models, starting with gpt-3.5-turbo-0125, provides greater customization and performance optimization.
Other changes to OpenAI API
Apart from the upgrades and changes we mentioned above, OpenAI recently introduced some other, not necessarily technical, changes to make the usage of their tool more flexible and cost-effective. There are three areas where new changes were recently introduced:
MORE COST-MANAGEMENT OPTIONS
OpenAI API is now more cost-effective, as Open AI recently introduced two more ways to allow users to reduce costs, both using OpenAI API consistently and irregularly
- Discounted committed throughput: Customers with a sustained token per minute (TPM) usage on GPT-4 or GPT-4 Turbo can now access provisioned throughput with discounts of up to 50% based on the commitment size.
- Batch API for asynchronous workloads: This new batch API allows users to run non-urgent workloads asynchronously, as they are now priced at 50% off shared prices. However, because this new feature returns results within 24 hours, it can be used for some less time-sensitive issues such as model evaluation, offline classification or summarization.
COMPREHENSIVE API DOCUMENTATION
Now, users have access to comprehensive API documentation that provides them with everything they need to use this tool effectively. This documentation includes:
- Guides
- Tutorials
- Best practices
Additionally, OpenAI uses its most experienced and knowledgeable employees to provide enterprise clients with high-quality customer service.
CONTINUOUS UPDATES
OpenAI is constantly working on its API and other products to ensure that all users get access to the latest advancements in AI. The same is true concerning OpenAI API – the company strives to provide users with frequent updates and new features, so if you’re using this tool, it is very likely that in a few months from now this tool will be even more flexible and comprehensive.
Wrapping up
If you use OpenAI API, we invite you to explore all the advanced features of this technology as they can allow your company to make the most of the full potential of artificial intelligence in a safe and effective way. And if that’s the first time you’re reading about OpenAI API, we have a comprehensive step-by-step guide for companies that want to start using this technology. Read more here: How Do I Use the OpenAI API and Find the API Key? Step by Step Guide.