Command A+

Cohere's open 218B enterprise model: 25B active, 48 languages, images and tools, Apache 2.0, runs on two H100s at 4-bit.

In short. Command A+ is Cohere's open model for businesses, built to handle documents, many languages and tool-using agents. It reads text and images. Its license is Apache 2.0, a big change from Cohere's earlier non-commercial releases. A compressed version runs on just two H100 GPUs.

Summary

Cohere released Command A+ on 20 May 2026 under the Apache 2.0 license. It is a sparse mixture-of-experts model with 218 billion parameters in total and 25 billion active per token. It reads text and images, writes text, and handles reasoning, tool use and retrieval-augmented tasks. Cohere says it supports 48 languages, up from 23 in the earlier Command A, and has a 128K-token input context with up to 64K tokens of output.

Cohere says it replaces several of its earlier models, Command A, Command A Reasoning and Command A Vision, in one. The company reports that, compared with Command A Reasoning, it improves τ²-Bench Telecom from 37% to 85% and Terminal-Bench Hard from 3% to 25%, and that it reaches 37 on the Artificial Analysis Intelligence Index. These are Cohere's own figures.

Weights come in three precisions: BF16 at about 438 GB, FP8 at about 225 GB and a 4-bit W4A4 build in NVFP4 format at about 132 GB. The model card lists a minimum of two H100 or one B200 GPU for the W4A4 build and recommends it for most uses, saying all three show negligible quality differences. It runs on vLLM and Transformers.

Earlier Cohere open-weight releases, such as Command A in March 2025, used the non-commercial CC BY-NC license. Command A+ is Cohere's move to a fully permissive license.

Good for

  • Enterprise agents, retrieval-augmented generation and document work with a clean Apache 2.0 license
  • Multilingual deployments covering 48 languages
  • Reading charts, scans and screenshots alongside text
  • Private deployment on two H100s or one Blackwell GPU using the 4-bit build

Watch out for

  • The 4-bit W4A4 build is in NVFP4 format and is aimed at Blackwell GPUs; on other hardware check which build your serving stack supports
  • It outputs text only and has no audio or video input
  • Benchmark gains over earlier Command models are Cohere's own measurements
  • The 128K context is shorter than the million-token windows of many 2026 flagship models

Performance claims come from the developer and have not been independently verified.

How much memory does it need?

Estimates for the Command A+ W4A4 variant. A mixture-of-experts model must hold all of its experts in memory even though only some are used for each token, so total parameters decide the memory and active parameters decide the speed.

Variants Parameters BF16 FP8 / INT8 4-bit (Q4) Fits on
Command A+ BF16 BF16 218B 25B active 501 GB 251 GB 150 GB Big-memory desktop
Command A+ FP8 FP8 218B 25B active 501 GB 251 GB 150 GB Big-memory desktop
Command A+ W4A4 NVFP4 218B 25B active 501 GB 251 GB 145 GB Big-memory desktop

Estimates: weights plus about 15% overhead. Real usage grows with context length and depends on the software.

VRAM calculator

The license in plain words

One of the most widely used permissive licenses. You may use, change and sell products built on the model with only light conditions.

More about Apache 2.0

See also

Frequently asked questions

What is Command A+?
Command A+ is Cohere's open model for businesses, built to handle documents, many languages and tool-using agents. It reads text and images. Its license is Apache 2.0, a big change from Cohere's earlier non-commercial releases. A compressed version runs on just two H100 GPUs.
Can I use Command A+ commercially?
Yes. Command A+ is released under Apache 2.0, a permissive license. You can use it commercially if you follow the standard conditions, such as keeping the license notice.
Is Command A+ open source?
Command A+ is released under Apache 2.0, a license approved by the Open Source Initiative. Training data and training code may not be public, so it is best described as open weights under an open-source license.
Can I run Command A+ on my own computer?
Only on a machine with a lot of memory. The smallest variant, Command A+ W4A4, needs about 145 GB at 4-bit precision. Desktops with 128 to 512 GB of unified memory can hold it, at modest speed.
When was Command A+ released?
Command A+ was released by Cohere on May 20, 2026.
How long a context window does Command A+ have?
Command A+ supports a context window of up to 128K tokens.

Official links

Sources

Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.

Release date and facts last verified November 8, 2026.

Related models

Model Released Parameters Context License Smallest variant needs
Mistral Large 4 Mistral AI 1.05T 52B active 1M Mistral Open-Weight (Conditional) 671 GB GPU server
Reflection Beam Reflection AI 501B 23.4B active 1M Apache 2.0 (Permissive) 323 GB Big-memory desktop
Inkling Thinking Machines Lab 276B – 975B 41B active 1M Apache 2.0 (Permissive) 188 GB Big-memory desktop
Cohere Transcribe Cohere 2.07B – Apache 2.0 (Permissive) 4.75 GB Laptop or phone