Optimized Vector Embeddings & Search

We’ve significantly optimized our vector embeddings and search capabilities to improve performance and efficiency while maintaining high-quality semantic matching.

Key improvements:

• All job post embeddings now use 768-dimensional half-precision (16-bit) vectors (embed_3sh), cutting bandwidth and compute costs without sacrificing accuracy.

• Embedding data in CSV downloads is now much smaller, reducing storage and processing needs.

• The previous embed_3s parameter has been replaced with embed_3sh for consistency.

These refinements make semantic search and matching lighter, faster, and just as effective, ensuring you get the best balance of performance and relevance. Based on our tests, we found that smaller, leaner embeddings deliver nearly identical results - just faster and with far less overhead. For full details, see the updated Vector Embeddings and Search API Documentation.

Related Docs