Command A+
Language modelOpen weightsCard summarizedOur fastest, most powerful open-source model for running high-performance enterprise agents with maximum efficiency.
LiveSource: Introducing Command A+ (launch post)
Use it via API
Where to call it
- Cohere API
command-a-plus-05-2026 - Microsoft Foundry / Azure
coherelabs-command-a-plus-05-2026-w4a4 - Self-host (weights)
CohereLabs/command-a-plus-05-2026-bf16
Good for
- Enterprise agents mixing vision and text inputs
- Reasoning, tool use and retrieval-augmented generation
- Multilingual tasks and translation across 48 languages
Overview
Command A+
Our fastest, most powerful open-source model for running high-performance enterprise agents with maximum efficiency.
- Cost
- $0.80 host
- Context
- 128K
- Cheapest host
- $0.80 / $3.20
- Max output
- 64K
- Knows up to
- —
- Released
- 20 May 2026
The last model in Cohere's Command A family and its first mixture-of-experts model: 218B total and 25B active parameters, open weights under Apache 2.0. One set of weights covers image input, reasoning, tool use, translation and agentic tasks in 48 languages. Cohere says it runs on as little as one B200 or two H100s at W4A4 and is faster than Command A Reasoning. launch post; changelog
command-a-plus-05-2026- Input / output per 1M
- $0.80 / $3.20 via Microsoft Foundry
- Context window
- 128K
- Max output
- 64K
- Knows up to
- —
- Reasoning
- Reasoning model; thinking can be enabled or disabled
- Weights
- Open · Apache 2.0
About this card. A model card is the maker's own account of a model: what it's for, how it was trained and tested, and where it falls short. This page condenses Cohere's Introducing Command A+ (launch post). Every claim carries its source page.