NEW Perplexity Update is INSANE! 🤯

New Perplexity AI models are insane. Perplexity just dropped two brand new AI models and they are not like anything you've seen before. These aren't chatbot updates. These aren't prompt upgrades. These are embedding models. The engine behind how AI actually finds and understands information. And Perplexity just made that engine way more powerful. We're talking birectional understanding, insane compression, top of every major benchmark on the planet right now. If you use AI for search, content automation, or building any kind of smart tool for your business, you need to know about this. It changes how AI retrieves information. It changes how accurate your AI tools are. And it changes what's possible for people building with AI right now. Stay with me because by the end of this video, you'll know exactly what these models do, why they matter, and how to use them. Let's go. Hey, if we haven't met already, I'm the digital avatar of Julian Goldie, CEO of SEO agency Goldie Agency. Whilst he's helping clients get more leads and customers, I'm here to help you get the latest AI updates. Julian Goldie reads every comment, so make sure you comment below and let him know what you think of this one. All right, Perplexity AI just

released two brand new embedding model families. They are called PLX embed v1 and pplx embed context v1. And the moment I saw the benchmark results, I had to make this video immediately because these things are ridiculous. But before I get into the models, let me make sure you actually understand what an embedding is. Because if you skip this part, nothing else I say is going to land properly. Think about how AI search works. You type in a question, the AI finds the right answer. But how does it know what the right answer is? It's not just matching keywords, it's matching meaning. Here's how. Every piece of text, every word, every sentence, every document gets converted into a long list of numbers. That list of numbers is called a vector. And that vector represents the meaning of the text. When two pieces of text mean similar things, their vectors are close together. When you search for something, the AI finds the vectors that are closest to your search. That's embedding based retrieval. That's how modern AI search works. Now, here's the problem with older models. They only read text in one direction, left to right, like reading a sentence, but never being allowed to go back and check what you just read. That makes it hard to fully understand meaning. The AI misses context. It pulls wrong results. It gets confused. Perplexity just fixed that.

And they fixed it in a way nobody else has done at this scale. So, let's talk about what they actually built. Perplexity released two model families. Let me break down what makes each one different. The first is pplx embed. This is your standard dense text embedding model. It's built for speed and accuracy in traditional search. The second is pplx embed context v1. This one is different. This model doesn't just understand individual chunks of text. It understands how chunks relate to the whole document. It holds the full context of what it's reading. This makes it the go-to model for rag pipelines. Rag stands for a retrieval augmented generation. That's when your AI needs to pull from a large database of documents before answering a question. If you're building any kind of AI assistant, knowledge base, or automated research tool, PPLX Embed Context V1 is the one you want. Both models come in two sizes. 0. 6 billion parameters, which is fast and lightweight, and 4 billion parameters, which trades some speed for a big jump in accuracy. You pick based on what your use case needs. Quick one before we go deeper. If you're watching this and thinking, "This is cool, but I don't know how to actually use any of this for my business. " That's exactly why the AI Profit Boardroom exists. Inside the AI Profit Boardroom, we take tools and updates like this and show you how to plug them into real business systems. We're already working on use cases around Perplexity's new embedding models, building smarter knowledge bases, better AI search tools, and faster automation workflows. If you want to learn how to use AI like this to run a smarter, faster, more automated business, the AI profit boardroom is where that happens. Link is in the description. Now, let's get back into

it. Here's the technical part. I'm going to keep it simple because it doesn't need to be complicated to be powerful. Most AI models use what's called a causal architecture. They read text from left to right and predict what comes next. That's how GPT models work. That's how most language models work. It's great for generating text, but it's not ideal for understanding meaning across a full document. Perplexity use something called diffusionbased pre-training. What this does is take a standard causal model. In this case, they used Quen 3 as the backbone and convert it into a birectional encoder. That means the model reads the full sequence at once. It sees the whole sentence, the whole paragraph, the whole chunk, all at the same time, both directions, left and right, simultaneously. Let's talk

numbers because the numbers here are serious. On MTE, which is one of the most respected public benchmarks for text embeddings in the world, Perplexity's new models are sitting at the top. on Contour AB which specifically tests contextual retrieval. PPLX embed context V1 at 4 billion parameters sets a brand new record. It beats everything else currently out there for contextual chunk level retrieval. On Bergen, another major retrieval benchmark. These models are leading again across the board. Perplexity has built the most accurate embedding models available right now for real world retrieval tasks. That's not marketing, that's benchmark data. And it matters because benchmark performance in this space translates directly to how accurate your AI tools are in production. Getting access is

straightforward. Perplexity has released these models on HuggingFace under the MIT license. That means they're open, free to use, free to build with. They're also available through the Perplexity API. So if you're already using Perplexity's platform, you can start experimenting with these immediately. They integrate with transformers, sentence transformers, and ONX. That means they plug cleanly into whatever AI stack you're already running. Lang chain, custom rag pipelines, whatever you've got, these models drop straight in. Here's the big picture. Most people and most businesses using AI right now are using tools with weaker retrieval underneath. Keyword search, generic embeddings, models that don't fully understand context. That's why AI tools still get things wrong. That's why chat bots still pull irrelevant information. That's why search inside AI tools still frustrates people. The businesses that are going to win with AI are the ones that build smarter systems. Systems where the AI actually understands the content it's searching through. Perplexity's new embedding models are a meaningful step toward that. Better retrieval means better answers. Better answers means better tools. Better tools means more value for whoever is using them. Perplexity dropped two new

embedding model families. pplx embed v1 for standard retrieval. PLX embed context V1 for contextual rag pipelines. Both come in 0. 6B and 4B sizes. They use birectional understanding through diffusion based pre-training. They use native int8 and binary quantization for smaller faster storage. And they're currently leading on MTeb conten benchmarks. They're live right now on hugging face and the perplexity API. If you want to learn how to use tools like this inside a real business system, join the AI profit boardroom. We take the best AI tools and updates and show you how to actually implement them. SOPs, workflows, real examples, real use cases. The link is in the description. And if you're just getting started and want something completely free, join the AI success lab. It's our free community with over 40,000 members all learning how to use AI properly. You'll get the full notes from this video, access to 100 plus AI use cases, templates, and a community of people who are actually doing the work. Links are in the comments in the description. Comment below and let Julian know, are you already using embedding models in your business, or is this the first time you've heard about this? He reads every single comment. See you in the next one. Word count dash 680