Update Cohere Node: Support (input_type) parameter for embeddings and add in cohere embedding v4 + reranker v4

Feature Requests
1

The idea is:

Update the Cohere node to support the latest v4 generation models and the mandatory input_type parameter.

  • New Models: Add support for rerank-v4.0 (pro/fast) and embed-multimodal-v4.
  • Input Type: Add the input_type dropdown (options: search_query, search_document, classification, clustering). This is required for optimal performance on v3/v4 models.
  • Future-proofing: Please allow manual string input for the model name field so we can use future models immediately without waiting for node updates.

My use case:

I am building advanced RAG agents. I specifically need the 32k context window provided by Rerank v4 to process larger document chunks. Without the input_type parameter, the embeddings for “search_query” vs “search_document” are not generated correctly, degrading retrieval quality.

I think it would be beneficial to add this because:

  • Performance: v4 models offer SOTA accuracy and speed (especially v4-fast).
  • Context Limit: The new 32k context window is a game-changer for enterprise RAG.
  • API Parity: The current node lags behind the API.

Any resources to support this?

Are you willing to work on this?

No