OCI Generative AI now supports Cohere Embed 4

OCI Generative AI has added support for Cohere Embed 4. This model processes mixed modality inputs (text and images in a single payload) to generate text embeddings.

Key Features
  • Matryoshka Embeddings
    • Generates embeddings in vector dimensions: 256, 512, 1,024, and 1,536.
    • Nested vector representations, not available in Embed 3 models.
  • Mixed Modality Input
    • Combine text and images in one payload for text embeddings.
    • Available through API only (not in the Console).
    • Images must be base64 encoded. Use a token calculator for image token counts (for example, 512x512 image ≈ 1,610 tokens).
  • Context Length
    • In the API and SDK, total tokens per embedding can be up to 128,000 tokens per run.
    • In the Console, text inputs < 512 tokens each and maximum 96 inputs per run.
  • Language Support
Available Regions
  • Brazil East (Sao Paulo) (dedicated AI cluster only)
  • Germany Central (Frankfurt) (dedicated AI cluster only)
  • Japan Central (Osaka)
  • UAE East (Dubai) (dedicated AI cluster only)
  • UK South (London) (dedicated AI cluster only)
  • US Midwest (Chicago)
Access

Available through chat interfaces, API, and dedicated endpoints, depending on region.

Resources