Embedding API gains contrastive pre-training for text and code embeddings

AI Impact Summary

New capability enables text and code embeddings produced via a contrastive pre-training objective, creating a shared latent space for natural language and code. This should improve semantic search accuracy and code intelligence tasks, reducing retrieval errors across text and code queries. Teams should compare the new embeddings against current models, adapt pipelines to consume the updated vectors, and re-index vector stores as needed to realize the performance gains.