OpenAI Reveals Latest Embedding Models and GPT Turbo Enhancements

OpenAI, a leading player in generative AI, has introduced innovative features for developers to manage and monitor API keys effectively. The company has also implemented updates for its GPT-3.5 Turbo and GPT-4 large language models, along with the introduction of two new text embedding models.

The rollout, executed on January 25, includes improvements to API key management aimed at offering developers enhanced visibility into usage patterns and increased control over API keys. Firstly, developers can now assign specific permissions to API keys directly from the API keys page. For instance, a key can be designated with read-only access for powering an internal tracking dashboard or restricted to access only specific endpoints.

Secondly, the API usage dashboard and export function now provide metrics for an API key after activation. This enables a straightforward analysis of usage on a per-team, feature, product, or project level by utilizing separate API keys for each purpose. OpenAI has plans for further enhancements to empower developers with more comprehensive views of API usage and key management.

OpenAI has unveiled two novel text embedding models: a “smaller and highly efficient” text-embedding-3-small model and a “larger and more powerful” text-embedding-3-large model. The latter is touted as OpenAI’s new top-performing embedding model.

In the realm of generative AI, embeddings are sequences of numbers representing concepts within content, be it code or natural language. OpenAI clarifies that embeddings facilitate the understanding of relationships between content, enabling machine learning models and algorithms to perform tasks such as retrieval or clustering. Applications like knowledge retrieval in ChatGPT or the Assistants API heavily rely on these embeddings.

The updated GPT-4 Turbo preview model, named gpt-4-0125-preview, demonstrates enhanced performance in tasks like code generation, addressing previous issues of incomplete task execution, or “laziness,” as described by OpenAI. This model also includes a fix for a bug affecting non-English UTF-8 generations. For users seeking automatic upgrades to new GPT-4 Turbo preview versions, OpenAI is introducing the gpt-4-turbo-preview model name alias, directing to the latest GPT-4 Turbo preview model.

Additionally, OpenAI is launching a new GPT-3.5 Turbo model, gpt-3.5-turbo-0125, accompanied by a reduction in prices for GPT-3.5 Turbo usage. This update enhances the model’s accuracy in responding to requested formats and addresses a bug related to text encoding for non-English language function calls.