OpenAI Embeddings API adds contrastive pre-training for text and code embeddings

AI Impact Summary

The capability introduces contrastive pre-training to generate embeddings for both text and code, delivering cross-modal semantic representations. This should improve retrieval quality and code search accuracy, enabling more effective similarity ranking and clustering for mixed queries. Downstream pipelines may need re-indexing and adjustments to embedding model selection, as the new space can shift results and costs. Validate with a pilot to benchmark latency, max token usage, and indexing pipelines against current embeddings.