NEW Gemini 3.0, Claude 4.5, New Qwen Models, & New OpenAI Models!

The last few weeks have been wild in the AI space, and this week is no different. We've got leaks about Google's Gemini 3. 0, rumors that Entropics Claude 4. 5 might be right around the corner, Alibaba dropping trillion parameter Quen models, and OpenAI teasing brand new features in the next couple of weeks. It feels like every lab is sprinting right now. And honestly, as users, we're the ones who win. So, let's break it all down in today's video. So, first up, Gemini 3. 0. know if you've been scrolling through X, you probably saw these leaks. Supposedly, Google already has Gemini 3. 0 Ultra Pro and Flash being tested internally. Users are also seeing AB testing prompts appear when they are using Gemini 2. 5. Remember, 2. 0 only dropped recently. Pro for reasoning task and flash for speed and low latency. So, to see 3. 0 0 pop up this quickly. That tells me that Google is on a 6-mon cycle and maybe even shorter. So now, what might we get with this 3. 0? There are some few guesses. Number one, reasoning upgrades. Ultra could compete head-to-head with GPT5 on complex prompts. We can also experience faster speeds. Flash might become the fastest model in the industry, and this is great for chat apps. Lastly, multimodality. They've been hyping up video and image understanding, so 3. 0 could double down on that. And let's not forget, Google has to deliver right now. Gemini 1. 5 got flack for being behind GPT4, and Gemini 2. 0 was better, but it didn't blow people away. So 3. 0 is their chance to actually leapfrog and challenge chat GPT. Next up, Entropic. There's this post floating around on X where Hattie Zho, who's actually a research scientist at Entropic, is confirming that Claude 4. 5 might be just around the corner. Now, Claude 4 only came out earlier this year with Opus for top performance, Sonnet as a balance option, and Haiku for lightweight task. So, 4. 5 feels like Entropic's way of keeping their lineup fresh. Why does this matter? Because Claude is already one of the most trusted models when it comes to writing and research. If you ever use Claude for long- form content, you know it's less likely to hallucinate and it keeps a more natural tone. So, if 4. 5 makes Sonet even smoother or if it improves factual accuracy, that's huge. Of course, this is still a rumor and Entropic hasn't confirmed anything on official websites, but with Google and OpenAI gearing up for big moves, it makes sense that Entropic wouldn't just sit back. They need to stay in the conversation as well to be relevant. All right, now let's talk about China's Alibaba, which brings Quen 3 Omni. This is the first open-source natively end-to-end Omni model that unifies text, image, audio, and video in one mode. So that means no modality trade-offs. Unlike earlier non-native multimodal models that bolted speech or vision onto text first models, Quen 3 Omni integrates all of the modalities from the start. So it allows it to process the inputs and generate outputs while maintaining realtime responsiveness. It also has state-of-the-art on 22 out of 36 audio and AV benchmarks. There's only 200 milliseconds latency, which makes the conversations, especially voice and video chats, feel instant and natural. It can also process and understand up to 30 minutes of audio at once. This allows you to ask questions about long recordings, meetings, or podcast. It's also fully customizable via system prompts. It has built-in tool calling and open-source captioner, which also has low hallucination. This model will unlock a very wide use cases. First off, realtime speech-to-pech assistance for customer support, tutoring or accessibility, cross language chat and voice translation across 100 plus languages, meeting transcription, summarization and captioning of audio, video, generating captions and descriptions for audio and video content tool integrated agents that can call APIs or external services. personalized AI characters or assistants with customizable styles and behaviors. Another really cool thing about Quen is that it's multilingual out of the box. It's super strong in Chinese, but also really solid in English, which makes it a global competitor. And unlike Google or OpenAI, Alibaba is releasing a lot of these models under Apache 2. 0 licensing. That means developers worldwide can build on top of them without being locked into closed systems. This matters because while Western labs fight for enterprise deals, Quen is becoming the go-to for open-source innovation. Imagine the next wave of AI startups. A lot of them might be powered by Quinn because of its open-source nature.

Another thing to note is that there are three distinct versions of Quen 3, Omni, 30B, and A3B. Each of these serve different purposes. The Instruct model is the most complete. It combines both the thinker and talker components to handle audio, video and text inputs and to generate both text and speech outputs. The thinking model focuses on reasoning task and long chain of thought processing. It accepts the same multimodal inputs but limits output to text making it more suitable for applications where detailed written responses are needed. And lastly, the captioner model is a fine-tuned variant built specifically for audio captioning, producing accurate, low hallucination text descriptions of audio inputs. And finally, OpenAI. Sam Alman has been teasing some compute inensive features rolling out to pro users in the next few weeks. He hasn't said exactly what they are, but given the timing, this could be early steps towards improving GPT5. They also just released GPT5 codecs which is tuned for coding. That's a pretty big hint. They're carving out specialized models again like Kodak used to be for GPT3. And then there's the Nvidia deal. 10 GW of compute. That's insane. That's more capacity than some countries use for AI. When you combine that with pro only feature testing, it's clear that they're gearing up for something huge. Maybe multimodal video, maybe memory that actually works across sessions, maybe GPT6 itself. Either way, the message is big things are coming soon. And what the hell is happening with all these deals? Why is OpenAI making a deal with Oracle and then Oracle is Nvidia and Nvidia OpenAI? Seems like an infinite money glitch. So, to recap, Google's Gemini 3. 0 finally might be in testing. Entropic could drop Claude 4. 5 any day now and Alibaba just flexed with a trillion parameter Quen model and Open AI is teasing something major in the next couple weeks. The AI arms race is in full swing and we're the ones along for the ride. Which of these are you most hyped about? Gemini, Claude, Quen, or OpenAI's next drop? Let me know down in the comments. And of course, hit subscribe if you want to stay on top of all of this because at the speed things are moving, you'll miss something big every week if you