Cohere Embed 5: One Index, Two Speeds

🔥 What happened Cohere dropped Embed 5: two embedding models (Pro and Fast) sharing one vector space. Index with Pro, query with Fast — no quality loss. Both handle text, images, and 128K tokens across 100+ languages. 💡 Why it matters Fast processes 377 docs/sec vs. Pro's 160 — at 98.4% of Pro's quality. That's the lever for agents firing dozens of searches per task. Storage drops from 819 GB to 3.2 GB across 100M chunks. ⚡ Our take Cohere built the first embedding model for the agent economy. But the benchmarks use Cohere's own RCP-nDCG@10 metric — independent replication is still pending.