Top AI Product

Cohere released Command A+ — a 218B-parameter sparse MoE model (25B active) under full Apache 2.0, the company’s first fully open-weight model. It’s tuned for complex reasoning, multimodal document processing, and agentic workflows, and it runs on a single NVIDIA B200 or just two H100s.

## Native citations

The standout feature: when Command A+ retrieves information from an external tool, it generates explicit “grounding spans” — special tags embedded in the output that link every factual claim to the specific source document or database row. For enterprise RAG, that’s the difference between “trust the model” and “verify every claim against its source.” Built into the model, not bolted on afterward.

## Sovereign-AI positioning

Apache 2.0 weights on Hugging Face is a deliberate bet on “sovereign AI” — enterprises, governments, and developers running frontier-grade models entirely inside their own secure environments. The tokenizer is optimized for 48 languages, cutting tokens for Arabic by 20%, Japanese 18%, Korean 16% — a direct play for non-Western enterprise markets where US-centric tokenizers waste budget.

## Why it matters

A genuinely deployable open frontier model (2 H100s, not a datacenter), plus native citations, plus permissive licensing, is a precise enterprise wedge. Cohere isn’t chasing chatbot benchmarks — it’s chasing the regulated, sovereignty-sensitive buyers that closed APIs can’t serve.