Command A+

Name: Command A+
Author: Cohere

by CohereCommand

APIAvailable through a hosted API — pay per token, no self-hosting required

128K context≈ 96,000 words218B (25B active) params

Command A+ is Cohere's flagship enterprise model, built around a 218B sparse MoE architecture that activates only 25B parameters per forward pass — efficient enough to run on two H100 GPUs. Its standout feature is native citation grounding: responses can reference source passages with inline citations, making it particularly suited to RAG pipelines and enterprise knowledge systems. With 48-language support and Apache 2.0 licensing, it targets organizations that need deployable, multilingual AI with verifiable outputs.

Capabilities

Capability scores are AI-generated based on model documentation, benchmarks, and technical specifications. Learn more

Tool Use

Exceptional

Instruction Following

Command A+

by CohereCommand

APIAvailable through a hosted API — pay per token, no self-hosting required

128K context≈ 96,000 words218B (25B active) params

Command A+ is Cohere's flagship enterprise model, built around a 218B sparse MoE architecture that activates only 25B parameters per forward pass — efficient enough to run on two H100 GPUs. Its standout feature is native citation grounding: responses can reference source passages with inline citations, making it particularly suited to RAG pipelines and enterprise knowledge systems. With 48-language support and Apache 2.0 licensing, it targets organizations that need deployable, multilingual AI with verifiable outputs.

Capabilities

Capability scores are AI-generated based on model documentation, benchmarks, and technical specifications. Learn more

Tool Use

Exceptional

Instruction Following

Glossary

ArchitectureThe underlying structural design of a neural network that defines how data flows through layers and components.Citation GroundingWhether an AI model's claims are supported by and properly attributed to specific sources.Forward PassA single computation cycle where input data flows through the model's layers to produce an output prediction.GroundingThe practice of ensuring a model's responses are based on and supported by provided source documents rather than generated from general knowledge.MultilingualA model trained to understand and generate text in multiple languages, not just English.ParametersThe learned numerical values in a model — more parameters generally means more capacity but higher compute cost.RAGRetrieval-Augmented Generation — a technique that grounds model responses in retrieved documents to improve accuracy.Sparse MoEA mixture-of-experts design where only a small fraction of the model's parameters are used for each prediction, reducing computational cost while maintaining model capacity.

Capabilities

Capabilities

Use Case Fit

Glossary