Sunday Jan 28 2024

OpenAI Launching new Embedding Models

In an ongoing effort to push the boundaries of machine learning and artificial intelligence, OpenAI has introduced a suite of new updates that are set to redefine how developers and organizations interact with AI technologies. Most notably, the launch of new embedding models alongside API updates is stirring anticipation and excitement within the tech community. Here's a detailed look into what OpenAI has in store with these latest advancements, including lower pricing on GPT-3.5 Turbo and an array of tools designed to streamline API usage management.

Revolutionizing Content Understanding with New Embedding Models

At the forefront of these releases are two innovative embedding models, designed to enhance how machines interpret and analyze content. These models, named text-embedding-3-small and text-embedding-3-large, represent breakthroughs in efficiency and performance for a wide range of applications, from knowledge retrieval to the enrichment of developer tools.

The Power of Text Embeddings

Text embeddings function as a sequence of numbers, encapsulating the concepts within content, whether it be natural language or code. This method allows AI models to grasp the relationships between different pieces of content, streamlining processes like clustering, retrieval, and more. OpenAI's embedding models are pivotal for knowledge retrieval in ChatGPT, Assistants API, and numerous Retrieval Augmented Generation (RAG) developer tools.

Small Yet Mighty: text-embedding-3-small

The highly efficient text-embedding-3-small model marks a significant upgrade from its predecessor, offering stronger performance across various benchmarks while reducing the associated costs. Specifically, it exhibits a notable increase in the average score on the multi-language retrieval (MIRACL) and English tasks (MTEB) benchmarks, all while being priced 5X lower than the previous generation.

Superior Performance: text-embedding-3-large

For those requiring more robust capabilities, the text-embedding-3-large model creates embeddings with up to 3072 dimensions. It stands as OpenAI's top-performing model, demonstrating impressive scores on both MIRACL and MTEB benchmarks. This model is designed for applications demanding the highest level of accuracy and conceptual understanding from their embeddings.

Flexibility with Shortened Embeddings

A noteworthy feature of these new models is the ability to shorten embeddings without losing their concept-representing properties. This flexibility allows developers to balance performance and cost-efficiency effectively, catering to specific needs such as storage constraints in vector data stores.

Advancements Beyond Embeddings

OpenAI's commitment to innovation is further underscored by updates to other models and pricing adjustments:

  • GPT-3.5 Turbo: An updated model with a 50% reduction in input prices and a 25% decrease in output prices, alongside improvements in accuracy and format response capabilities.

  • GPT-4 Turbo Preview: An enhanced version offering more thorough task completions and solutions to previous bugs affecting non-English UTF-8 generations.

  • Moderation Model: The release of text-moderation-007 presents the most robust effort yet in identifying potentially harmful text, bolstering the safety and reliability of AI applications.

Empowering Developers with Enhanced API Management

Recognizing the importance of seamless integration, OpenAI introduces new tools for API management. Developers can now assign permissions to API keys and gain detailed insights into API usage on a per-key basis. These improvements aim to provide developers with greater control and visibility over their AI systems, paving the way for more efficient and secure AI applications.

Looking Ahead

As OpenAI continues to lead in the development of cutting-edge AI technologies, these updates highlight a steadfast commitment to making AI more accessible, efficient, and powerful. With these new embedding models and API enhancements, developers and organizations are equipped with the tools needed to unlock new possibilities and drive innovation in their field. For the latest on OpenAI's APIs and further advancements, the tech community is encouraged to stay connected and follow OpenAI's developments closely.


This stride in AI innovation is made possible by the collective efforts of OpenAI's team, including notable contributions from researchers and developers dedicated to pushing the envelope of what's possible in AI. Their relentless pursuit of excellence continues to shape the future of technology and artificial intelligence.

Stay tuned to OpenAI's blog for more updates on research, product announcements, and developments in the world of AI.