Cohere cracks lossless quantization and native citations with its first fully Apache 2.0-licensed open model, Command A+

Cohere launched Command A+, a 218B sparse MoE enterprise model for reasoning, multimodal document work, and agent workflows.

The model weights are available on Hugging Face under Apache 2.0, with native citations built in for traceability.

Cohere says selective 4-bit quantization and distillation preserve quality while cutting compute to one Blackwell B200 or two H100 GPUs.

The open, multilingual model gives enterprises and governments a more controllable frontier AI option with lower infrastructure demands.