​Cohere has introduced two new models on Microsoft Azure AI Foundry to enhance Retrieval-Augmented Generation (RAG) and agentic AI workflows

1. Embed 4: Multimodal and Multilingual Embedding Model for RAG and Semantic Search

Key features:

• Supports over 100 languages for text embeddings, enabling cross-lingual search capabilities.​

• Image embedding support, allowing for multimodal search scenarios.​AIbase

• Matryoshka embeddings with scalable dimensions (e.g., 256, 512, 1024, 1536), providing flexibility between accuracy and resource usage.​

• Int8 quantization and binary embedding output, reducing storage requirements and enhancing search speed.​

2. Command A: Instruction-Tuned Conversational LLM Designed for Enterprise AI Scenarios

Key features:

• 256K token context window, enabling processing of extensive documents in a single prompt.​

• Excels in instructions, summarization, and RAG workflows, with built-in tool calling capabilities.​

• Supports all major business languages, including Japanese, Korean, and German.​

• 150% higher throughput compared to the previous generation, resulting in higher performance and lower latency.​

Deployment and Integration via Azure AI Foundry

Azure AI Foundry offers straightforward deployment options and serverless API endpoints for utilizing Cohere models. It also supports integration with Azure AI Agent Service, facilitating the construction of complex agentic AI workflows.​

Potential Applications

These models can be applied in various scenarios, including:

• Financial report summarization

• Legal research assistance

• Technical knowledge assistance

• Multimodal search and indexing