NEW DeepSeek V3.2 Update Will Blow Your Mind! 🤯

Deepseek just dropped something crazy. They cut their API prices by 50%. And they added new tech that makes long chats way faster. I'm going to show you exactly what changed, how to use it right now. Plus, I'll run live tests so you can see the real results. Stay until the end because this could save you thousands on AI costs. Deepseek released version 3. 2 experimental. This is their follow-up to version 3. 1. They added something called Deepseek Sparse Attention. Stick with me because this changes everything. 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. Normal AI models look at every single word when they process text. That takes a lot of computer power. It's slow and it costs money. Deepseek sparse attention is different. It only looks at the important words. Think of it like reading a book. You skim some parts and focus on others. That's what this does. When you use AI for long conversations or long documents, it gets expensive fast. This new way keeps costs low, even with really long text, and it's just as accurate for most tasks. Deepseek cut their API prices by more than 50%. If you were paying $100 a month, you now pay 50. If you were paying 1,000, you now pay 500. This adds up fast if you're running a business. Let me break down who should care about this. First, developers building apps that use long context like chat bots or agents. You get the same quality for half the price. Second, researchers who want to test sparse attention. This is open source on hugging face. Third, product makers using AI in your products. Lower costs mean better margins. There are three ways to use this. You can use the Deepseek web app, API, or you can download the model from HuggingFace and run it yourself. The web app is easiest. Go to the Deepseek website, sign in, start chatting. The API is for developers. You write code that talks to Deepseek servers. The API pricing is where you see the 50% cut. Hugging face is for people who want full control. You download the model and run it on your own hardware. Let me show you a real example. I have a long meeting transcript, 6,000 words. I'm going to ask Deepseek to summarize it and create an action list with names and deadlines. Here's the prompt. summarize this 6,000word meeting and produce an action list with assignees and deadlines. I paste the transcript. I hit enter. It processes the whole thing in seconds. It gives me a clean summary. Then it lists every action item. Who's responsible? And when it's due, this would take a human an hour. It took the AI 10 seconds. The cost difference is huge. With old pricing, this costs 2 cents. With new pricing, it costs 1 cent. Multiply that by 10,000 requests. Old price is $200. New price is 100. That's real money. All right, we're about 2 minutes in and I want to tell you about something cool. We've got a flash sale on the AI profit boardroom. It's 30% off this month. Plus, you get free access to the SEO elite circle when you sign up. This deal lasts through September. Only a few days left. The AI Profit Boardroom is the best place to scale your business, get more customers, and save hundreds with AI automation. Now, back to the tech. Let me explain how sparse attention works. Imagine you have a sentence with 100 words. Normal attention looks at every pair of words. That's 100 * 100 calculations. 10,000 calculations. Sparse attention picks 20 important words. Then it only looks at those. That's 20 * 20 calculations. 400 instead of 10,000. Way faster, way cheaper. The benchmarks show that version 3. 2 performs about the same as version 3. 1 on most tasks. It's close enough that most people won't notice a difference. Here's the trade-off. If you need perfect accuracy for something critical like medical or legal work, test it carefully. If you're doing normal stuff like chat bots or summaries, you're fine. This is why it's called experimental. Let me show you another demo. This time I'm using the API. I wrote a simple Python script. It calls the DeepSeek API. It sends a long document. It asks for a summary. The model name is Deepseek v3. 2 experimental. It comes back in 4 seconds. The summary is solid and the cost is half of what version 3. 1 would cost. This is huge for anyone building AI products. Customer support bots can serve more customers without raising prices. Content tools can process more articles for the same budget. Research can analyze bigger data sets. Deepseek released this on hugging face. That means you can download it and run it locally. People are testing it on H200 GPUs, MI350 from AMD. Even some neural processing units. The community is building support patches on day one. If you want to try it, go to HuggingFace, search for Deepseek V3 3. 2 Experimental. You'll see the model card with instructions. You need a decent GPU. If you don't have one, you can rent one from a cloud service. One thing to watch out for, this is experimental. It might have bugs. It might change. Don't build your entire business on it until it's

stable. But do test it. Sparse attention is the future. Let me give you some use cases. Long document analysis. You can throw in 50page reports and get summaries. Multi-turn conversations. chat bots that remember everything, agent systems, AI that can do multiple tasks in sequence, code generation, give it a long code base and ask it to add features. All of these get better with longer context and all get cheaper with sparse attention. Deepseek is competing with OpenAI, Google, Anthropic, but Deepseek is way cheaper. This price cuts even more pressure on them. We might see a price wall. That's good for us. If you're already using Deepseek, it's easy to switch. You just change the model name in your API calls. If you're new to DeepSeek, you need to sign up, get an API key, then you can start making requests. This is version 3. 2 experimental. That means there will be a stable version later. Deepseek is testing this in the wild. They want feedback. If you use it now, you're shaping the future of the product. Here's my verdict. If you need long context and low cost, try version 3. 2 experimental. If you need maximum accuracy for sensitive tasks, compare it to version 3. 1 first. But for most people, this is a no-brainer. Same quality, half the price, faster speed, action steps. First, go to hugging face and check out the model page. Second, sign up for the Deep Seek API if you haven't already. Third, run a test. Pick a real task from your business. Try it on 3. 2. See if it works. This is experimental. Test it thoroughly. Make sure it works for your needs. If you notice quality drops, switch back to 3. 1. You can use both. Pick the right tool for each job. This price cut is a big deal. It opens doors. If you've been thinking about building an AI tool, but the cost scared you, now you can. If you've been using expensive APIs and struggling to make the numbers work, you can switch and save. The key is to take action. Don't just watch this video and move on. Go try it. Pick one thing you want to automate. Use Deep Seek to do it. See what happens. Want to make more money with AI? Welcome to the free AI money lab with Julian Goldie. Inside, you'll get 50 plus free AI tools and 200 plus chat GPT SEO prompts. You'll learn how to make money with AI agents, get 10,000 plus free NAT workflows, 200 plus chat GPT prompts, a full blueprint to generate thousands of leads free with AI. What you'll also get is a free AI community, free AI course, and proven AI case studies. Now, let me tell you about the AI profit boardroom. Again, we've got a flash sale going on. It's 30% off this month. You get free access to the SEO Elite Circle when you sign up. The deal lasts through September. Only a few days left. This is the best place to scale your business, get more customers, and save hundreds with AI automation. Go to the huginface model page. Try DeepseekV 3. 2 experimental. Tell me in the comments which real world task I should benchmark next. Customer support, content creation, code generation. Let me know. If you got value from this, smash that like button. Subscribe if you haven't already. The AI is moving fast and I'm here to keep you in the loop.