Cohere Command A+: The First Apache 2.0 Licensed Frontier Open Model

Command A+ — What Makes It Special?

Cohere has released Command A+, a 218-billion-parameter language model that represents a major leap in open-source AI. Co-founded by Aidan Gomez, co-author of the landmark "Attention Is All You Need" paper, Cohere built Command A+ for complex reasoning, multimodal document processing, and agentic workflows.

The most groundbreaking aspect isn't just the model's capabilities — it's the license. Command A+ is released under Apache 2.0, one of the most permissive open-source licenses available. This means enterprises, governments, and developers can run, modify, and deploy the model commercially with virtually no restrictions.

Sparse Architecture That Runs Lean

Command A+ uses a Sparse Mixture-of-Experts (MoE) architecture. While it houses 218 billion total parameters, only 25 billion are active during any generation step. This makes it dramatically cheaper to run than proprietary models like GPT-5.5 or Claude Opus 4.7, which are estimated to be in the trillions of parameters.

The model comes in three formats: 16-bit (BF16), 8-bit (FP8), and a highly compressed 4-bit (W4A4). Cohere claims the W4A4 quantization is lossless — meaning the compressed version performs identically to the full-precision model, a technical achievement that could change how enterprises think about model deployment costs.

Why This Matters for Builders

If you're building AI-powered products, Command A+ gives you frontier-grade capabilities without vendor lock-in or per-token API costs. You can run it on your own infrastructure, fine-tune it for your use case, and keep all data in-house. The Apache 2.0 license means no usage restrictions, no revenue sharing, and no surprise terms of service changes.

Common Questions (FAQ)

Q1: Can I run Command A+ on a single GPU? A1: The 4-bit W4A4 version significantly reduces hardware requirements, but 218B parameters still need substantial GPU memory. Check Hugging Face for specific hardware recommendations.

Q2: How does it compare to GPT-5.5? A2: It's competitive on reasoning and agentic benchmarks while being far cheaper to run, though GPT-5.5 may edge ahead on some specialized tasks.

Q3: Where can I download it? A3: The model weights are available on Hugging Face under CohereLabs/command-a-plus-05-2026-w4a4.

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