Founded in 2019 by Transformer co-author Aidan Gomez, Nick Frosst, and Ivan Zhang, Cohere released Embed v3, introducing compression-aware training that preserves retrieval accuracy under int8 and binary vector Quantization while pairing dense Embeddings with cross-encoder Rerank scoring for production RAG pipelines.
Cohere
Also known as: Cohere Labs · Cohere For AI · Command R · Command A · Aya · Cohere Rerank
Plain English
An enterprise AI lab founded in Toronto in 2019 by Transformer co-author Aidan Gomez, Nick Frosst, and Ivan Zhang, specializing in retrieval-augmented generation (RAG), neural search (Embed and Rerank), enterprise agentic models (Command R and Command A), and the Aya multilingual open-science initiative.
Technical Overview
Cohere designs its stack around enterprise RAG, sovereign cloud/on-prem deployment, and multilingual coverage: dense compression-aware Embeddings (Embed) pair with cross-encoder passage rerankers (Rerank) to feed citation-grounded generation and multi-step tool use in the Command family (Command R+, Command A, and the Apache-2.0 sparse MoE Command A+), while its research arm Cohere Labs (formerly Cohere For AI) develops the Aya and Aya Vision multilingual open-research models across 100+ languages.
Timeline (6)
2023
Release2024
PaperCohere's non-profit research division Cohere For AI (now Cohere Labs) published Aya 101 and the open Aya Dataset, coordinating over 3,000 independent researchers across 119 countries to build human-annotated instruction-tuning corpora and an Apache-2.0 multilingual model spanning 101 languages (over half of them lower-resourced).
- Release
Following the March 2024 debut of Command R (35B), Cohere launched the 104B-parameter Command R+ across Microsoft Azure, Amazon Bedrock, and Hugging Face with a 128K context window, native inline document citation generation for RAG, and multi-step tool use across 10 key enterprise languages.
2025
ReleaseCohere launched Command A, a 111B-parameter model with a 256K-token context window. Cohere reported that the model can run on two A100 or H100 GPUs.
2026
ReleaseCohere released Command A+, its first flagship enterprise model published under a permissive Apache-2.0 license—combining a 218B-total / 25B-active sparse Mixture-of-Experts architecture with unified reasoning, vision, 48-language fluency, and multi-step agentic tool execution for private on-prem deployment.
- Release
Cohere launched Rerank 5 (in Pro and Fast tiers), adding Context Focus—which extracts exact relevant character spans from retrieved documents before passing context to downstream LLMs—alongside calibrated relevance scoring across 100+ languages to cut RAG token overhead and latency.
See also (16)
Open full graphCohere
11 more connections
- OpenAIOutgoing · Alternative to
- Tool Use & Function CallingOutgoing · Enables
- Amazon & AWS AIIncoming · Uses
- Cohere Command & AyaOutgoing · MentionsIncoming · Created by
- US AI LabsIncoming · Enables
- Context Window & RoPEOutgoing · Mentions
- Mixture of Experts (MoE)Outgoing · Mentions
- Open-Weights LicensingOutgoing · Mentions
- QuantizationOutgoing · Mentions
- Reasoning & Test-Time ComputeOutgoing · Mentions
- TransformerOutgoing · Mentions