Command A+

Cohere's open-weight multimodal mixture-of-experts model for vision, reasoning, translation, and agentic tasks, with 218B total and 25B active parameters.

Running this yourself: likely needs a high-memory cloud gpu.