# AI News: DeepSeek V4, GLM 5, NEW Gemini Agent, Grok Code

## Метаданные

- **Канал:** Universe of AI
- **YouTube:** https://www.youtube.com/watch?v=HaWhG5CytD8
- **Дата:** 09.01.2026
- **Длительность:** 10:07
- **Просмотры:** 11,785
- **Источник:** https://ekstraktznaniy.ru/video/8919

## Описание

AI news is moving fast. In this video, we break down DeepSeek V4’s upcoming coding-focused model, GLM-5 following Z.ai’s IPO, Google’s new Gemini AI agent inside Gmail, and xAI’s push into Grok Code for developers.

For hands-on demos, tools, workflows, and dev-focused content, check out World of AI, our channel dedicated to building with these models:  ‪‪ ⁨‪‪‪‪‪‪‪@intheworldofai 

🔗 My Links:
📩 Sponsor a Video or Feature Your Product: intheuniverseofaiz@gmail.com
🔥 Become a Patron (Private Discord): /worldofai
🧠 Follow me on Twitter: https://x.com/UniverseofAIz
🌐 Website: https://www.worldzofai.com
🚨 Subscribe To The FREE AI Newsletter For Regular AI Updates: https://intheworldofai.com/

ai news, artificial intelligence news, ai weekly news, ai model news, ai updates, latest ai news, llm news, ai industry news, ai research news, frontier ai models, ai labs, deepseek v4 news, deepseek ai update, glm 5 news, z ai glm, gemini ai news, gemini agent update, google gemini ai news, gmail ai 

## Транскрипт

### Intro []

If you think AI progress has slowed down, well, it hasn't. It just shifted direction. This week, Deepseek is preparing version 4, targeting serious coding workloads. GLM5 is loading as China's next flagship model. Google quietly shipped a Gemini agent. XAI is improving Grock's coding abilities with the new model. This isn't about smarter chat pods anymore. It's about who owns real work.

### DeepSeek V4 [0:25]

So, let's break it down. According to Routers, Chinese AI startup Deepseek is expected to launch its next generation model version 4 in midFebruary. What makes this interesting is the focus. Version 4 is being positioned as a coding first model and internal testing by Deepseek employees suggests it could outperform rivals like Entropics Claude and Open AI's GPT models specifically on coding tasks. One of the biggest reported breakthrough is versions 4 ability to handle extremely long coding prompts which could be a real advantage for developers working with large complex code bases. Now this hasn't been independently verified yet and Deepseek hasn't commented publicly but the context here matters. Deepseek has quietly become one of the most important players in China's AI push, especially as the country works to build its own AI ecosystem and reduce reliance on foreign chips. The company already drew global attention after Silicon Valley executives praised earlier models like Deepseek version 3 and Deepseek R1. At the same time, Deepseek has been criticized in some countries over security and privacy concerns. So, version 4 isn't just about performance. It's also about credibility on the global stage. If these claims hold up, Deep Seek version 4 could be one of the most serious coding focused model launches we've seen this year. Just to show you guys where Deep Seek currently stands versus Claude Opus 4. 5. I've asked both these model to make a Voxil style world, add trees and a village, but also include surprise elements. So, what we're going to look at are two worlds created by Deepseek version 3. 2. 2 and Opus 4. 5. And the reason why I'm showing you guys this is that you can clearly notice from this demo that version 4 has to cover a big gap to not only match clot opus 4. 5, but take it over. So this is the Voxer world created by Deepseek. And we can see it's not bad. Like we see our villages, we see our trees. It's pretty blocky. Uh kind of reminds me of Roblox for some reason. But that's what the voxal world looks like in DeSeek versions. and you can move in, you can move your mouse and everything like that. Now, let's take a look at Claude Opus 4. 5. Now, from the get- go, you can see Claude Opus 4. 5 looks much better when it comes to design elements. It added in animals. I guess the surprise element was I can find diamonds, which is pretty cool. A tiny dragon up here. So, like it's adding cool things like that. So, this is amazing as well. And like you can see if I click on this, more stuff is happening. So, this is a huge gap that Deepseek has to cover. So, if employees are saying that the new version 4 is going to be able to not only compete with Opus 4. 5, but be even better than that, then I can't imagine what's cooking and I'm really excited to test out this new model. Now, let's talk

### GLM 5 [3:22]

about Z. AI because this is a big moment, especially for AI and companies. Z. AI AI has officially gone public on the Hong Kong Stock Exchange under stock code 02513. That's not just a headline. It changes the game. Going public means Z. AI now has real capital, public market pressure, and a much bigger mandate to ship. This isn't a research lab experimenting anymore. This is a company that now has to execute on a big scale. And the timing matters. Chinese AI companies are entering public markets right as global competition is heating up, which means models aren't just competing on quality, they're competing on speed, pricing, and scale. Now, here's where it gets interesting. ZAI didn't just announce the IPO, they also shared what's coming next. They said a new free model is shipping as early as next week, along with a brand new image model and expanded support like PayPal. But the most important line is this. GLM5 is currently in training. That tells us two things. First, they're using the IPO momentum to accelerate product releases. Second, GLM5 isn't vaporware. It's already deep in development. Why is this huge? Because public companies can't afford slow iteration. Every release now matters to investors, developers, and enterprise customers. And here's the bigger strategic angle. ZAI has historically leaned into open and accessible models, especially compared to fully closed western labs. If GLM5 continues that direction, whether through open weights, aggressive pricing, or generous free tiers, it puts real pressure on open AI, Google, and entropic. Not just on performance, but on business models. Open-source and lowcost models don't need to be the best. They just need to be good enough and widely adopted. So, this IPO isn't just about funding. It's about turning ZAI into a serious global competitor with the capital to train Frontier models, the incentive to ship fast, and the leverage to disrupt pricing across the entire AI market. If GLM5 lands strong, this could be one of the most important model launches of the

### Grok Code [5:46]

year. Now, let's talk about XAI because Grock is quietly ramping up in a serious way. A lot of people are watching for Grock 4. 2 to be the next major model release and Elon Musk has already hinted that a meaningful Grock code update is coming soon with more improvements expected over the next few months. What's interesting is that we're now getting an early look at how Grock code is actually going to work. Based on recent findings, XAI is introducing something called Grock Build. Essentially, a dedicated coding interface inside the web version of Grock. Instead of everything living in a chat window, Grock build appears as a separate tab with a central prompt bar and a more structured setup designed specifically for building. Users will be able to choose between different agents. And while we don't have every detail yet, the direction is clear. This is a developer first update. Inside the interface, there are controls for things like setting up environments, linking repositories, managing workspace paths, handling environment variables and secrets, and even controlling internet and script access. There's also built-in GitHub support, which is a big signal. What really stands out though is how Grock Code is launching. The first version is expected to focus on local agents. That means you install a package, run Grock from your command line, and connect it locally rather than everything happening in the cloud. Remote environments are marked as coming soon. That tells us XAI is prioritizing control, flexibility, and a secure local execution, which is exactly what more advanced developers care about. This isn't Grock trying to be the flashiest coding assistant on day one. It's Grock trying to become a modular customizable tool that serious users can actually build with. And if you zoom out, this fits XAI's broader strategy. They're not just chasing benchmarks. They're building systems that iterate fast, integrate deeply, and can expand over time, especially once remote agents and upgraded Grock code versions come online. So, while Grock may not be the top coding model today, this setup makes one thing clear. XAI is laying the groundwork to compete seriously in developer workflows and not just conversations. Over 3 billion people use

### Gemini Agent [8:09]

Gmail every day. And while AI has been part of Gmail for years, email itself has changed a lot since 2004. Inbox volume is at an all-time high and managing information has become just as hard as sending emails. So, Google is officially bringing Gmail into the Gemini era, turning your inbox into a proactive AI assistant. The biggest change is something called AI overviews. Instead of searching and digging through old threads, Gmail can now summarize long email conversations automatically and answer questions about your inbox in plain English. You can ask things like, "Who was the plumber that sent me a quote last year? " or "Find renovation estimates from around March. " Gmail handles the reasoning and gives you a direct answer. No scrolling required. Conversation summaries are rolling out to everyone for free. Inbox Q& A is available to Google AI Pro and Ultra subscribers. Google is also upgrading how you write emails. Help me write is now available to everyone and suggested replies are smarter using full conversation context to draft responses in your tone. There's also a new proofread feature for deeper grammar and style checks for paid users. And finally, Google is testing AI inbox, which highlights what actually matters, like bills, appointments, and important messages while pushing all the noise down. All of this is powered by Gemini 3 and is rolling out starting today in the US. The takeaway is simple. This isn't Gmail adding AI features. This is Gmail becoming an AI agent. Instead of working inside your inbox, your inbox starts working for you. Make sure to subscribe

### Outro [9:49]

to our channel. We do real tests, not just headlines. Make sure you're also subscribed to The World of AI. And don't forget to check out our newsletter for deeper breakdowns you won't see on YouTube. And I'm growing my Twitter following, so make sure you follow me on Twitter as well. Hope you guys enjoyed today's video and I'll see you in the next