Cohere and Perplexity Ship New Enterprise Embedding Models

Two major releases this week advance enterprise search and retrieval capabilities.

Key points:

• Cohere Embed 5 ships in Pro and Fast tiers that share the exact same vector space • This allows indexing documents once with Pro and running everyday searches with the cheaper Fast model • Perplexity open-sourced pplx-embed-v2-context-9b-preview, a new contextual embedding model • Both releases target cutting vector-database costs for enterprise retrieval

The shared vector space approach is particularly notable — it addresses a common pain point where switching between embedding model tiers previously required re-indexing entire document libraries.

If your team maintains a search or retrieval system, test re-indexing a sample document set with Embed 5 Pro and querying it with Embed 5 Fast to evaluate the real-world cost and speed difference.

Why It Matters: The shared vector space between Cohere's Pro and Fast tiers eliminates the costly re-indexing typically required when switching between embedding model tiers, directly reducing operational overhead for enterprise search systems.