Sorry for not submitting a PR, just wanted to document this!

First issue:

• In BedrockEmbeddings support for Cohere Embed v4 #680, the issue describes that the Langchain-AWS batch embedding functions for Cohere models didn't support the new output schema of Cohere Embed v4.
• This was fixed in Cohere Embed v4 schema #681, but that commit did not apply the same fix to the non-batch embedding function for Cohere models. See: https://github.com/langchain-ai/langchain-aws/blob/main/libs/aws/langchain_aws/embeddings/bedrock.py#L192
• Right now users can't use Cohere Embed v4 for normal non-batch inputs

Second issue:

• In https://github.com/langchain-ai/langchain-aws/blob/main/libs/aws/langchain_aws/embeddings/bedrock.py#L357, Cohere embedding models have their input sizes limited to no more than 2048 characters, as expected by Cohere Embed v3.
• Cohere Embed v4 seem to support a maximum context length of 128K tokens for each text input. That's a significant increase from v3, and allows for the embedding of entire documents without requiring custom chunking logic. See: https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-embed-v4.html
• Right now users can't use that extended input length support because the old v3 limit still applies.