In the artificial intelligence, OpenAI continues to lead the charge with groundbreaking advancements. The recent release of new embedding models and API updates marks another significant stride in OpenAI's commitment to enhancing the capabilities of AI developers. In this blog post, we'll delve into the details of the latest innovations, including two new embedding models, updates to GPT-3.5 Turbo, and improved ways for developers to manage API keys.

► New Embedding Models: A Leap Forward in Efficiency

Embeddings, as sequences of numbers representing content concepts, play a crucial role in enabling machine learning models to grasp relationships between content elements. OpenAI introduces two new embedding models, each catering to different needs:

1. text-embedding-3-small:

This highly efficient model supersedes its predecessor, text-embedding-ada-002, delivering stronger performance. Benchmarked against multi-language retrieval and English tasks, text-embedding-3-small showcases a remarkable increase in scores, making it a cost-effective and high-performing choice for developers.

    • Stronger Performance:

Scores on multi-language retrieval (MIRACL) have surged from 31.4% to 44.0%, and English tasks (MTEB) scores have risen from 61.0% to 62.3%.

    • Reduced Price:

Achieving a 5X efficiency boost, text-embedding-3-small boasts a reduced price from $0.0001 to $0.00002 per 1k tokens compared to text-embedding-ada-002.

2. text-embedding-3-large:

This next-generation, larger embedding model generates embeddings with up to 3072 dimensions, offering unparalleled performance.

    • Enhanced Performance:

Compared to text-embedding-ada-002, text-embedding-3-large showcases remarkable improvements, with MIRACL scores increasing from 31.4% to 54.9% and MTEB scores increasing from 61.0% to 64.6%.

    • Native Support for Shortening Embeddings:

Recognizing the trade-off between larger embeddings and computational resources, OpenAI introduces a technique to shorten embeddings without compromising their conceptual representation, providing flexibility to developers.

► Updated GPT-3.5 Turbo Model and Lower Pricing

OpenAI's commitment to customer scalability is evident with the introduction of the new GPT-3.5 Turbo model, gpt-3.5-turbo-0125. With a 50% reduction in input prices and a 25% reduction in output prices, this model is poised to deliver higher accuracy in requested formats and addresses a text encoding issue for non-English language function calls.

Developers utilizing the pinned gpt-3.5-turbo model alias will be seamlessly upgraded to gpt-3.5-turbo-0125 two weeks after its launch.

► Updated GPT-4 Turbo Preview Model

The GPT-4 Turbo preview model, gpt-4-0125-preview, reflects OpenAI's dedication to refining the model for optimal performance. With a focus on thorough completion of tasks like code generation and improvements in non-English UTF-8 generations, this model sets the stage for the upcoming launch of GPT-4 Turbo with vision in general availability.

► Updated Moderation Model for Enhanced Safety

In the pursuit of fostering a safer online environment, OpenAI introduces text-moderation-007, the most robust moderation model to date. With continuous improvements in safety, developers can utilize the free Moderation API to identify potentially harmful text, furthering OpenAI's commitment to safety best practices.

► Enhanced Visibility and Control with New Platform Improvements

OpenAI introduces two platform improvements aimed at empowering developers with enhanced visibility and control over API keys:

1. Key Permissions:

Developers can now assign specific permissions to API keys directly from the API keys page, providing flexibility for various use cases, such as read-only access for internal tracking dashboards or restricted access to specific endpoints.

2. Usage Dashboard Metrics:

The usage dashboard and export function now expose metrics on an API key level after turning on tracking. This enhancement enables developers to gain insights into usage on a per-feature, team, product, or project basis by utilizing separate API keys for each.

 

OpenAI's latest embedding models and API updates underscore the organization's commitment to fostering innovation, efficiency, and safety in the realm of artificial intelligence. As developers explore the enhanced capabilities of the new models and leverage improved API management tools, the future of AI development looks increasingly promising. OpenAI continues to pave the way for transformative advancements, setting the stage for a dynamic and evolving AI landscape.