On 20 May 2026, Cohere announced Command A+. Cohere released its larger enterprise model under Apache 2.0 with vision input, tool use and 48-language support.
This release brief was checked against first-party material on 10 August 2026. The date above is the public announcement date, not the date a repository was created or a third-party provider added the model. Where access or weights arrived later, that distinction is recorded below.
Release record
| Field | Verified detail |
|---|---|
| Announcement | 20 May 2026 |
| Availability or weight release | 20 May 2026 through Cohere and downloadable Hugging Face weights |
| Release type | open-weight multilingual multimodal foundation model |
| Access | Apache 2.0 weights in BF16, FP8 and W4A4 plus hosted endpoints |
| Architecture | a 218B sparse MoE with 25B active parameters and 8 of 128 routed experts per token |
| Maximum stated context | 128K input and up to 64K output |
What changed
Command A+ combined a commercially familiar licence with a model that can fit on one B200 or two H100-class accelerators in quantised form. Cohere also published multiple official quantisations and reported negligible benchmark differences among them, which gives deployment teams a clearer path to compare memory, speed and quality. Its value proposition is therefore controllable enterprise deployment rather than a single top leaderboard rank.
The practical comparison is therefore not simply whether Command A+ has the largest headline score. Teams need to ask whether its architecture, access terms, latency, tool behaviour and evaluation setup match the workload they actually intend to run. A model can lead one harness while losing on cost, refusal behaviour, multilingual quality or repeatability in another.
Benchmarks worth retaining
| Evaluation | Reported result | How to read it |
|---|---|---|
| Active parameters | 25B of 218B total | Sparse inference footprint, not a quality score |
| Official quantisations | BF16, FP8 and W4A4 | Cohere reports negligible benchmark-quality differences |
| Language coverage | 48 languages | Training coverage that still requires task-specific evaluation |
These are release-time results, not independently reproduced guarantees. The model card’s headline comparison is strongest when read with the exact quantisation and Cohere chat template; third-party serving stacks may change tool parsing and multilingual behaviour. Scores should remain attached to the disclosed effort setting, agent harness, tool access, timeout, context-management policy and judge model. Moving a number into a procurement sheet without those conditions creates false comparability.
Architecture and access
Command A+ is described as a 218B sparse MoE with 25B active parameters and 8 of 128 routed experts per token with 128K input and up to 64K output of stated context. Its access position at verification time is Apache 2.0 weights in BF16, FP8 and W4A4 plus hosted endpoints. That wording matters: open weights, source-available weights, an API, a product preview and a research demonstration give adopters very different rights and different levels of reproducibility.
Before deployment, record the exact model identifier or checkpoint, inference stack, quantisation, reasoning setting, region, price schedule and supplier terms. If the release uses a custom licence, read the licence itself rather than relying on the word “open” in launch copy. If it is API-only, preserve the dated documentation and change-notice route because the served snapshot can change without a downloadable artefact.
What an evaluation should test next
For Command A+, a credible internal gate should include:
- a frozen set of representative tasks with pass, fail and abstain criteria;
- a matched baseline using the same tools, timeout, prompt budget and reviewer rubric;
- repeated runs to expose variance rather than reporting a single best attempt;
- latency, token use and total task cost alongside task success;
- adversarial, multilingual and long-context cases relevant to the real deployment; and
- rollback evidence showing the previous model can be restored safely.
The wider model change-control guide explains how to keep model, prompt, tool and corpus changes reconstructable. The AI dependency inventory guide covers the release and supplier records needed after deployment.
AIEngine verdict
Command A+ is one of the window’s most practical open-weight enterprise releases. The Apache licence and official low-precision variants reduce adoption friction, but teams should still benchmark their languages, documents and tool schemas before replacing a hosted model.
This is a launch assessment, not a certification. Benchmark leadership is useful evidence of where to test; it is not authorization to place the model in a high-impact workflow without domain evaluation, security review and an accountable owner.
Primary sources
Image provenance
Hero image: Cohere official release artwork. The locally served WebP is a crop of the first-party release or model-card asset recorded in the repository provenance manifest.
