Cohere's Command A+: The First Apache 2.0 Open Model With Native Citations

Cohere just dropped something the AI community has been waiting for: a high-performance open model with a truly permissive license. Command A+ is a 218-billion-parameter model released under Apache 2.0 — and it comes with features that enterprise teams have been desperately needing.

What Makes Command A+ Special?

Two breakthroughs set this model apart. First, lossless quantization: the model runs efficiently without any degradation in output quality. Second, native citations — the model embeds special tags in its output that link every factual claim directly to the source document or database row it pulled from.

Why Apache 2.0 Matters

Most "open" AI models come with restrictive licenses that limit commercial use. Apache 2.0 is one of the most permissive licenses available, meaning any company can use, modify, and deploy Command A+ commercially without legal concerns. This is a first for a model of this caliber.

Use Cases — From RAG to Agentic Workflows

Command A+ excels at complex reasoning, multimodal document processing, and agentic workflows. The native citation feature makes it particularly valuable for regulated industries like finance, healthcare, and legal — anywhere you need to verify where an AI got its information.

Common Questions (FAQ)

Q: How does Command A+ compare to GPT-4 and Claude? A: It's competitive on reasoning benchmarks while offering the advantage of self-hosting and full data privacy.

Q: Where can I access the model weights? A: The weights are available on Hugging Face under the CohereLabs organization.

Q: What hardware do I need to run it? A: With lossless quantization, it runs on significantly less hardware than you'd expect for a 218B model. Exact specs depend on throughput requirements.

Stay ahead of the AI curve. Follow @AiForSuccess for daily insights.