XNews

Ada Reports 77ms Retrieval Latency Using Cohere Embed on Model Vault

Ada achieved 77 ms median retrieval latency, 2.7x greater throughput-per-dollar for offline embedding, a 4x smaller vector storage footprint with no measurable retrieval degradation, and 90.7% top-10 recall by using Cohere Embed on Model Vault. The results support faster, more accurate AI voice interactions and more cost-efficient information retrieval.

Event version 1.

Key insights

Low retrieval latency

Ada recorded 77 ms median retrieval latency using Cohere Embed on Model Vault. The latency supports faster, more accurate AI voice interactions.

Higher embedding throughput per dollar

Offline embedding delivered 2.7x greater throughput-per-dollar. The gain increases cost efficiency.

Smaller vector storage

The vector storage footprint was 4x smaller with no measurable retrieval degradation.

Strong retrieval recall

Ada reported 90.7% top-10 recall for accurate information retrieval.

Source posts