Command A+
218B-parameter, 25B-active MoE enterprise model released under Apache 2.0, deployable on-premises or in customer-controlled environments.
Score calculation
(7 + 7) ÷ 2 = 7.0
2 of six dimensions rated. Unrated dimensions and BDS status are excluded.
Full scoring method ↗Setup details
Model specifications/license are release-specific. Vendor's enterprise privacy/security posture does not establish every customer's deployment configuration.
Cohere announced Command A+ on May 20, 2026 and lists model ID command-a-plus-05-2026 under Apache 2.0. Checked 2026-10-23.
Energy
↑Cohere says Command A+ can run on one B200 or two H100s and promotes efficiency, but publishes no energy, water, or emissions per training run or task. Hardware minimum is not a measured footprint.
Reasoning & sources
Cohere publishes the hardware footprint needed for deployment and promotes efficiency, but not measured model energy or emissions. Hardware requirements bound a possible deployment; actual footprint still depends on utilization, electricity mix, and workload.
No model-specific energy measurement or independent validation found.
Training rights
↑The release and changelog provide technical specifications and license but no itemized training corpus or rights/consent register for Command A+.
Reasoning & sources
The Command A+ release and changelog describe model features and licensing but not the training corpus or consent process. Those technical details do not establish source permissions; request provenance and opt-out information.
Training-data provenance, licenses, and opt-out handling remain undisclosed in sources checked.
Privacy
↑Cohere promotes private deployment and says its models can run in a customer's VPC, on-premises, or air-gapped environment. Actual privacy depends on selected hosting and telemetry configuration; this is not a blanket API retention guarantee.
Reasoning & sources
Cohere supports VPC, on-premises, and air-gapped deployment, which can reduce exposure to an external host when configured by the customer. Those options do not prove isolation by themselves; assess logging, access, and retention in the selected deployment.
The release supports private, on-premises and air-gapped deployment, subject to operator configuration.
Need service-specific retention and training-use terms for API and private deployment.
Openness
↑Cohere says Command A+ is available under Apache 2.0 and offers model weights; its complete pretraining data and training process are not shown as fully reproducible in the release page.
Reasoning & sources
Cohere offers weights under Apache 2.0, giving broad reuse and local deployment rights. Because the training data and full recipe remain undisclosed, this is open-weight rather than fully reproducible openness.
Apache 2.0 weights and self-deployment are documented; the training corpus and full reproducible pipeline are not disclosed.
Open weights/license do not establish open training data or reproducibility.
Worker treatment
↑Cohere says Command A+ training used proprietary datasets with human annotation, automated and synthetic data, plus specialist data vendors. It does not disclose annotator pay, working conditions, or protections.
Reasoning & sources
Cohere’s exact Command A+ documentation confirms both human annotation and vendor-sourced datasets, so labor is part of this model’s pipeline. Because that disclosure omits compensation and protections, it supports asking about supplier standards but not a conclusion about worker treatment.
Annotator compensation, working conditions, exposure controls, vendor audits and current outcomes remain undisclosed in the reviewed documentation.
Government & military
↑Cohere says its MOU with defense-focused Indra includes Cohere sovereign LLMs for IndraMind and a defense cooperation line for mission-planning and multinational operations. The announcement does not name Command A+ as the deployed model or identify a completed contract.
Reasoning & sources
Cohere’s May 2026 announcement links its LLMs to IndraMind and describes defense cooperation around analysis, mission planning, and interoperability. This is a documented company-level defense relationship, but not proof that Command A+ itself is in use; the model-specific link remains unestablished.
The MOU does not establish a completed procurement, Command A+ deployment, operational use or independent oversight of the proposed defense work.
BDS status
↑We did not find Cohere among the named targets in the two official BNC campaign lists. This is a limited list check, not BDS approval.
Reasoning & sources
Cohere is not a named target in the checked lists, but BDS uses selective campaigns, so absence cannot establish a clean record. Running downloaded weights locally is a different decision from paying the developer or a cloud host; check the host as well.
This review covers the named targets in the BNC consumer/pressure guide and No Tech campaign, not every BDS campaign, local group, investment or supplier. Lists and ownership can change. A company’s absence from these lists is not an endorsement or proof that it has no relevant relationships.
All sources
- Introducing Command A+ ↗
- Cohere release notes ↗
- Command A+ for sovereign AI ↗
- An overview of Cohere models ↗
- Cohere Command A+ model data and training documentation ↗
- Cohere strategic MOU with Indra Group including defense cooperation ↗
- BNC: live guide to consumer boycott and pressure targets ↗
- BNC: No Tech for Oppression, Apartheid or Genocide campaign ↗