This NEW Cohere Command A+ is a GAME CHANGER!🤯

追加: 2026-05-26

[00:00:00]This new Cohere Command A+ is insane, and I'm going to show you exactly why it matters for your business right now. May 20th, Cohere dropped something big, really big. Model called Command A+ 218 billion parameters, and it runs on just two H100 GPUs. If you don't know what that means yet, you will by the end of this. Because what Cohere just did changes who gets access to Frontier AI, and that includes you. Let me start with the number that stopped everyone in their tracks, 218 billion parameters.

[00:00:29]That's the size of this model. Here's the thing, it only uses 25 billion of those parameters at any one time. That sounds weird, right? How do you have 218 billion, but only use 25 billion? Here's how it works. Imagine a hospital, massive hospital with every specialist you can think of. Cardiologist, a neurologist, a surgeon, a radiologist, hundreds of them. When you walk in with a broken arm, you don't activate the whole hospital. You get sent to the right specialist, right expert, stays on standby. That's exactly how Command A+ works. It's called a mixture of experts model, or MOE. The full 218 billion parameters are all there. Depending on what you ask it, only the relevant 25 billion activate. And that one decision, that single architectural choice, is why this model can run on just two H100 GPUs instead of a whole data center. That's the story nobody's telling loud enough.

[00:01:18]Because right now, if you want a frontier level AI model, the kind that competes with the best in the world, you usually need serious hardware. We're talking racks of GPUs, infrastructure that only the biggest companies on the planet can we haven't met already, I'm the digital avatar of Julian Goldie, CEO of SEO agency Goldie Agency.

[00:01:37]Whilst he's helping clients get more leads and customers, I'm here to help you get the latest AI updates. And Julian Goldie reads every comment, so make sure you comment below. And here's where it gets really interesting. The model isn't just small for its performance level, it's also fully open.

[00:01:51]Apache 2.0 license. That means businesses, governments, individuals, anyone can download it, run it, build with it, and use it commercially. No restrictions. No strings attached.

[00:02:01]Cohere co-founder Nick Frost put it plainly. He said, "This tech can go one of two ways. Can empower the people that use it. We are working towards that second one. That's the whole point." Now look, if you're building an AI-powered business or trying to automate and grow using AI, this kind of open access changes everything. Because you're not locked into one provider. You're not dependent on an API. You can run this yourself or have a team deploy it for you, and suddenly you've got frontier-grade AI working directly for your business. Inside the AI Profit Boardroom, we're already digging into exactly how tools like Command A Plus fit into a real business automation stack. We've got four live AI coaching calls every week where we go deep on models like this, how they compare, when to use them, and how to build workflows around them that save you hours every week. There are step-by-step daily tutorials, 30-day roadmaps built around the latest AI tools, and 2,800 business owners in the community right now. Many of them already experimenting with open-source models like this one. If you want to know how to actually use Command A Plus in your business, not just hear about it, link is in the description and comments, or go to aiprofitboardroom.com.

[00:03:05]Now back to the model. Cuz there's one more technical thing you need to hear about, and I promise it's worth it. It's called W4A4 quantization.

[00:03:13]Sounds complicated. Here's what it actually means. Normally AI models are stored in a format that takes up a lot of space and needs a lot of memory to run. Think of it like a full-resolution raw photo file. Detailed. Hard to move around.

[00:03:25]W4A4 compresses that file dramatically.

[00:03:27]Four-bit weights, four-bit activations.

[00:03:30]Basically, they found a way to squish the model down to a fraction of its original size without losing almost any quality. Cohere is calling it lossless quantization. They're saying the performance barely drops even after compression. And that compression is exactly why this model can run on two H100 GPUs instead of eight or 16.

[00:03:46]VentureBeat called it the technical centerpiece of this release. I'd agree.

[00:03:50]Now Cohere isn't claiming this beats every model on every benchmark. Some coding and general intelligence tests, it still trails models like DeepSeek-V4-Pro.

[00:03:58]That's not the point. The point is what you get for the hardware you're running on. Math and reasoning benchmarks, it's competing directly with models that are far bigger and more expensive to run.

[00:04:09]And A+ ranked first on something called the AA Omniscience Non-Hallucination Benchmark, 86% about three percentage points ahead of the next best model.

[00:04:18]That hallucination number matters, especially for businesses. Because here's what nobody talks about enough.

[00:04:23]When AI gets things wrong, when it makes stuff up, it creates problems. In a business context, that's a real issue.

[00:04:29]So, the fact that Command A+ ranked first for accuracy and not making things up, that's a big deal for anyone using AI to actually run operations. Let me give you a real example of how this plays out. Say you want to use Command A+ to do research on the best AI automation communities for business owners. You feed it a bunch of sources, articles, forum threads, community reviews. With native citations, it doesn't just give you an answer. It tells you which source backed up each claim. So, if you were, say, building a comparison document to show why the iProfit Boardroom is the right choice for business owners looking to automate with AI, you could do that in minutes with every claim sourced. That's a content asset that used to take hours to build. Now, let's talk about who this is actually for. Cohere built this with enterprises in mind. They've been very clear about something called sovereign AI. That's the idea that governments, businesses, and institutions should be able to run their own AI on their own infrastructure without handing all their data to a third-party API. Fujitsu, one of the biggest tech companies in Japan, already commented publicly that Command A+ is architecture aligns directly with their sovereign AI strategy. It's not just a model. It's a statement about where enterprise AI is heading. The model also expanded from 23 languages to 48. That includes all EU official languages plus major improvements in Arabic, Korean, and Japanese. And it's now multimodal, meaning it can process images and documents, not just text. It performed well on document and image reasoning benchmarks including Math Vista and MMMU. So, you've got a model that's efficient, open, multimodal, multilingual, citation-native, and runs on hardware that's actually accessible.

[00:06:00]That combination is rare. That's what makes this launch different. Here's the bigger picture though. For the last couple of years, the story in AI has been bigger is better. More parameters, more compute, more cost. And yes, the frontier models from the biggest labs are incredible. But what Cohere is proving with Command R Plus is that efficient can compete with enormous. The MoE architecture that only activates a fraction of parameters, took a 218 billion parameter model, figured out how to make it run on two GPUs without killing performance, made it fully open-source, baked in citations to reduce hallucinations, and shipped it with multimodal support and 48 languages. It's a piece of work. And it's available right now on Hugging Face. You can download it today. The acceleration in this space is real. A year ago, running a model anywhere near this capability level required infrastructure that only the biggest players could access. Now, it's two H100s and an Apache 2.0 license. That compression in time and cost and access, that's what you should be paying attention to. Because the question isn't whether AI is going to be part of how serious businesses operate. The question is how fast you're getting ready for it.

[00:07:02]Command R Plus is one more piece of evidence that the tools are getting better, cheaper, and more accessible faster than most people expect. And the business owners who are actually learning how to use these tools, not just reading about them but building with them, are the ones who going to look back in 18 months and be glad they started now. If you want to be inside a community that's tracking all of this and showing you exactly how to apply it in your business, how to pick the right model for the right task, how to build automation workflows that actually hold up, how to use tools like Command R Plus to save time and reach more customers, the AI Profit Boardroom is where that happens. Four coaching calls a week, daily tutorials on the tools that are actually shipping, 30-day roadmaps you can follow from day one, and 2,800 members who are doing exactly what you're trying to do. We've got people in there right now who are using open-source models in their business and we'll have resources specifically around Command A Plus as more use cases emerge.

[00:07:52]Link in the description, link in the comments, or go to aprofitboardroom.com.

[00:07:56]And if you want the full breakdown, the video notes, the use case library, all 100 plus AI workflows and SOPs, join the AI Success Lab. It's free. Links are in the comments and description. 67,000 members in there already. You'll get the notes from this video, templates you can use right now, and access to a community that's crushing it with AI every single day. The tools are moving fast. Command A Plus is proof. The question is whether you're moving with them.