Cohere Embed 5 lands: Pro and Fast models share one embedding space

Index with Pro, query on Fast — no re-embedding. Plus 85.8 on ViDoRe V3, ComfyUI gets a production API, and Codex gates whole-repo scans.

Nowline OCT 1 7:00 PM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • One embedding space, two speeds

    Embed 5 ships Pro for max-quality indexing and Fast for cheap, low-latency queries — and the two share a vector space, so you index once with Pro and serve live queries on Fast with no re-embedding. Both handle 128K context, 100+ languages, and Matryoshka dimensions from 256 to 2048.

  • RAG that reads tables and charts

    Pro posts the top average on ViDoRe V3 at 85.8, up from Embed 4's 77.0, and takes mixed text-and-image inputs — so retrieval works on visually dense PDFs without a separate parsing pass. Text runs $0.12/M on Pro and $0.08/M on Fast; images are $0.40/M on both.

  • ComfyUI gets a production API

    Comfy API turns a ComfyUI graph into an autoscaling endpoint with pinned custom nodes and versions that scales to zero when idle — a direct path from a local workflow to a shippable image or video service without re-implementing it yourself.

  • Codex tightens repo security scans

    Codex Security Cloud now runs OpenAI's Daybreak Blue by default for whole-repo vulnerability scans, and from Oct 1 individual users need an eligible security key to use it. If you lean on Codex for security review, check your access before your next scan.