North 2 登場：妥協なき AI を実現
アップデート内容を詳しく見るModels Overview
Cohere's generative and search models are trained, developed, and deployed by Cohere as trusted infrastructure for retrieval-augmented generation, semantic search, and enterprise AI applications. Our multilingual capabilities, security, and deployment flexibility provide the trust and capabilities enterprises need to build an efficient AI operation.
Generative Models
Frontier language models for complex reasoning, generation, and agentic workflows at enterprise scale.
Command A+
Our fastest, most powerful open-source model for running high-performance enterprise agents with maximum efficiency.
Ultra-efficient serving
Multilingual across 48 languages
Tool use and reasoning
Apache 2.0
Command A Vision
A multimodal enterprise model designed to extract, interpret, and reason over visual and document-based information within complex business workflows.
Multilingual image understanding
128K context window
Privately deployable
Apache 2.0
Command R7B
An efficient enterprise language model for reasoning, retrieval-augmented generation, and scalable private deployment in production environments.
Supports tool use
128K token context window
Runs on commodity GPUs and edge devices
CC-BY-NC-4.0
North Small Translate
North Small Translate delivers the highest quality open machine translation — setting a new standard for open translation research and development.
Multilingual across 50+ languages
98.8% terminology accuracy
Designed for cost efficiency
CC BY-NC 4.0
Transcribe
A high-accuracy speech recognition model for fast, reliable transcription across real-world enterprise audio and conversational workflows.
Supports 14 languages
25MB max file size
Batch processing
Apache 2.0
North Mini Code
Small, efficient, and open-source — our first agentic coding model, built for the sovereign developer ecosystem.
Agentic coding model
30B total; 3B active parameter efficiency
Privately deployable
Apache 2.0
North Micro Vision
A compact, customizable vision-language model for building high-performance image and document understanding applications.
2.4B parameters
Native-resolution visual processing
Multilingual and multi-image support
Apache 2.0
Command A Translate
An industry-leading translation model for accurate, context-aware multilingual communication across global customer, content, and localization workflows.
Supports 23 languages
Privately deployable
Tool use and structured outputs
CC-BY-NC-4.0
Search Models
The contextual foundation for accurate, relevant, and scalable enterprise AI.
Embed 5 Pro
The new industry standard for enterprise embeddings. Use for maximum retrieval quality across visually rich, financial, multilingual, and multimodal enterprise data.
State-of-the-art retrieval
128K context window
Multimodal text, image, and fused inputs
Multilingual across 100+ languages
Embed 5 Fast
A lightweight embedding model for latency-sensitive retrieval that outperforms models several times its size.
Same context and supported modalities as Pro
Leads compact models on document retrieval
Competitive performance at lower cost
Shared embedding space with Pro
Rerank 4 Fast
A low-latency reranking model for fast, scalable retrieval pipelines where speed and efficiency are critical.
32,768 context window
Multilingual across 100+ languages
High performance, lowest latency
Proprietary
Rerank 4 Pro
A high-performance reranker for maximum retrieval accuracy and nuanced relevance ranking in enterprise-grade search and agentic systems.
32,768 context window
Multilingual across 100+ languages
State of the art performance with low latency
Proprietary
Embed 4
Our fourth generation embedding model that for high-accuracy enterprise retrieval, semantic search, and RAG applications across complex real-world data.
128K context window
Multimodal (text, images)
Multilingual across 100+ languages
Proprietary
Parse 5
A high-fidelity vision parser that transforms enterprise documents into structured content for search, retrieval, and document intelligence workflows.
OCR and visual reasoning
Understands tables, forms, and handwriting
Structured outputs and bounding boxes
Optimal price-performance balance
Aya Models
A family of open multilingual models designed for global accessibility and research.
Aya Expanse
State-of-the-art multilingual performance for enterprise tasks spanning diverse languages, contexts, and enterprise environments.
32B parameters
23 languages
CC-BY-NC-4.0
Aya Vision
A multimodal language model for seamless communication across both text and images, enabling globally accessible AI interactions beyond language alone.
32B parameters
23 languages
Dynamic image captioning
CC-BY-NC-4.0
Tiny Aya – Global
A compact multilingual model for efficient local deployment, delivering strong translation and multilingual performance directly on-device without cloud dependence.
3.35B parameters
70 languages
Best balance across language and region
CC-BY-NC-4.0
Tiny Aya – Earth
Tiny Aya, optimized for West Asian and African languages.
3.35B parameters
70 languages
8K context length
CC-BY-NC-4.0
Tiny Aya – Fire
Tiny Aya, optimized for South Asian languages.
3.35B parameters
70 languages
8K context length
CC-BY-NC-4.0
Tiny Aya – Water
Tiny Aya, optimized for European and Asia-Pacific languages.
3.35B parameters
70 languages
8K context length
CC-BY-NC-4.0
