DeepSeek V3 is a free, open-source AI model that's 50-100x faster than the earlier R1 version and handles code, websites, and animations reliably. It has 671 billion parameters, a 128k-token context window, and an MIT license. You can run it locally or on cloud GPUs.
March 26, 2025
671 billion parameters
128k tokens
MIT (permissive) license
Approximately 50–100× faster
Today we are kicking off an absolute AI revolution. Goodness. This is the new version of Deep Seek and I think this might be it. R1 was their thinking AI, but this one is different. It is called V3 and the newest version is. Wow, it is truly something else. First, it is not a reasoning AI. Is that a good thing? Well, allow me to demonstrate it. Let's ask the earlier reasoning AI deepseek R1, what is the capital of France? And it is thinking and thinking and still thinking. Goodness. And then of course it gets it right. However, when we ask the new V3 look, we get an answer instantly. Yes, it is a touch less intelligent than its reasoning version theoretically. But in practice I am not so sure of that. You will see in this video that it might actually be even a bit better. Also, in return, it is perhaps fifty to a hundred times faster than the reasoning R depending on the question. That also means that you can run it cheaper. So let that sink in. This can hang with the best closed AI techniques out there. But wait, this is fully open and free for everyone that you can try online right now. The link is in the description. Now note that I myself like downloading models like this and running them myself privately. And my choice for that is Lambda, if you want to do that too. The instructions are in the video description. But now our question is. Okay, okay. These are benchmarks, numbers on a paper. Can it actually do all the cool stuff the clothes thinking AIs are capable of? Let's have a look. For instance, if it can write up a cool game easily, I would say that. Wow, that is exactly what we are seeing here. So what else can it do? Well look, it can code up this website in one shot. More than 800 lines of code super quickly. Now let's make it a bit harder. Like writing a sound wave visualizer. Another problem, we all did that ourselves in college by hand and it took at the very least hours of work and programming experience. And here you get it for free nearly instantly. Just pops into existence. That is mind blowing. Wow. And these are not just some one off results. It can write up really cool animations and all you need is just one text prompt to make them come alive. And there are more good news. Dear fellow scholars, this is two minute papers with Dr. Zsolnai. Oh yes, it has an MIT license. That is the D.O. whatever you want license. So it is as permissive as it gets. You can even ask for an interactive water molecule simulation where you can play with the temperature and see how the kinetic energy in the system changes accordingly. Now reasoning AIs like DeepSeq R1 can get you better results if you have super complex tasks. But most people don't need it. Most people Instead need this 1v3 which is much cheaper and faster. And get this in some cases even better too. Here it provides a little smoother motion than its rivals. Man, so good. And it is not just some isolated case. There are genuine use cases where it smokes the much more expensive reasoning model. What a time to be alive and yummy. We have a paper too. We fellow scholars are paper people so we love that. Especially that they showcase a needle in a haystack test. What is that? Well, this tells us about the model's precision when recalling information after seeing a lot of data. And it is all green baby. That is incredible. Even after reading a long document. 128k tokens. Let's say that is 250 to 300 pages of text and it recalls the details accurately. Now note that this model is still huge so you will probably need a bit of grant to run it. However, I am absolutely sure that this is going to kick off a huge AI revolution. This is exactly what we need and I feel that going forward, cutting edge AI systems are not going to be closed anymore. They will be free and increasingly more open for all of us. Think of all the good that we can do with these. Absolutely amazing. So what do you think? What would you fellow scholars use this for? Let me know in the comments below. Here you see me running the Full deep seq AI model through Lambda GPU cloud. The full 671 billion parameters running super fast and super reliably. This is insane. I love it and I use it on a regular basis. Lambda provides you with powerful Nvidia GPUs to run, run your own chatbots and experiments and it's the best. Seriously. Try it out now@lambdalabs.com papers or click the link in the description below.
