AI provider / Cohere
Cohere
A provider strongly associated with enterprise AI, retrieval workflows, multilingual coverage and private deployment options.
Cohere is the enterprise-first AI company, building model lines for reasoning, retrieval and multilingual work with secure, private deployment in mind. Its Command family rebuilt itself around RAG and tool use, turning retrieval-grounded answers into the core of the product.
From Command R and R+ to the two-GPU Command A and the 218B sparse Command A+, the line balances frontier capability with deployment that enterprises can actually control. Multilingual coverage and governance, not just benchmarks, define its pitch.
Model catalog
Every release in this provider’s model line.
Each row is a distinct model release — its date, its type and the traits it is known for. Open a source to dig into any of them.
| Model | Released | Type | Key traits | Source |
|---|---|---|---|---|
| North 2 | Oct 2026 | Agent platform |
| |
| Embed 5 | Sep 2026 | Embeddings |
| |
| Command A+ | May 2026 | Flagship |
| |
| Command A Reasoning/Vision/Translate | Aug 2025 | Model family |
| |
| Command A | Mar 2025 | Flagship |
| |
| Command R7B | Dec 2024 | Compact |
| |
| Command R/R+ refresh | Aug 2024 | Version update |
| |
| Command R+ | Apr 2024 | Flagship |
| |
| Command R | Mar 2024 | Version update |
| |
| Command | Nov 2022 | Debut |
|
Timeline
A practical timeline of model releases and company milestones.
This page tracks the moments that matter most: major model releases, version jumps and company moves that changed how this provider shows up in the AI market.
North 2 unifies Cohere’s agent platform around security and cost control.
Cross-session memory, org-wide shared agents and spend caps by user and team turn North into a full-stack agentic platform deployable in any cloud, on premises or fully air-gapped.
Embed 5 pushes retrieval quality on messy enterprise data.
Pro and Fast tiers share one embedding space — index with Pro, query with Fast — with multimodal input and a 128K context window aimed at financial filings, scanned PDFs and code.
Command A+ ships as a 218B sparse MoE under Apache 2.0.
Cohere’s biggest model yet pairs 25B active parameters with an open license, an unusually open move for the enterprise specialist.
Command A Reasoning, Vision and Translate specialize the flagship.
Three task-focused editions in one month turn Command A into a platform rather than a single model.
Command A becomes the flagship with a two-GPU footprint.
111B parameters, 23 languages and agent tooling, deployable on just two GPUs — efficiency as the headline feature.
Command R7B compresses the R recipe for cheap deployment.
The smallest R-family model brings RAG and tool use to commodity GPUs and even CPUs for high-volume use.
The 08-2024 refresh re-tunes R and R+ for throughput.
Around 50% higher throughput and 20% lower latency keep the family competitive without a headline version bump.
Command R+ scales the RAG-first recipe to 104B.
Multi-step tool use and ten business languages position it directly against GPT-4 Turbo on enterprise workloads.
Command R rebuilds the line around RAG and tool use.
A 35B model with 128K context and citation-grounded retrieval defines Cohere’s enterprise-search identity.
The first Command generation ships for enterprise text tasks.
Before the ChatGPT wave, Cohere is already selling instruction-following generation models to businesses via API.