Name: Kimi AI
Uploaded: 2025-02-23T12:03:51.627Z
Duration: 2 min 33 s
Description: AI assistant with ultra-long memory, capable of deep reasoning and web browsing.

kimi-ai

AI assistant with ultra-long memory, capable of deep reasoning and web browsing.

Kimi AI

Hi welcome to another video. So every day in my comments I'm asked to test a new model called Kimy K 1.5. Well I thought that today I'll do just that. 

Now what's Kim K 1.5? Well it's a model and by the way it's not open source but they have shared the technical report and stuff. We can also use it for free without any rate limits on their site as well. It actually launched I believe on the same day as R1. 

Anyway they say that K 1.5 is an 01 level multimodal model. So right off the bat we know that it is a multimodal model which makes it better than R1 in this sense as R1 can't look at images and stuff like that. It has short Chain of Thought performance outperforming GPT 40 and Claude Sonet 3.5 on AIM Math 500 and Live code bench by a large margin. 

It also has long Chain of Thought performance and matches.

01 across multiple modalities including math Vista Amy Code forces Etc. It is also a reasoning model if you didn't quite get that. And it's also trained with RL similar to R1. You can have a look at the benchmarks here where it beats 01 by a lower margin while also beating Claude and more, which is great to see. 

In training, they have used long context scaling, which basically means that in their reinforcement learning approach, they upped the context limit to 128k, which led to better training. They also have an improved policy optimization where they use the formulation of RL with long chain of thought and employ a variant of online mirror descent for robust policy optimization. 

It is also multimodal and is jointly trained on text and vision data, which gives it the capability of jointly reasoning over the two modalities.

The model is not open source as I had said. But you can try it out on their platform and it's free there. Without any rate limits, you'll find two sets of models on their platform. There's the base Kimi model which is simple and doesn't have Chain of Thought. And there's also the long thinking mode which makes it think. 

So I'll be testing the long think mode. Let's test it out. These are the 13 questions that I'm going to test it against. The first question is tell me the name of a country whose name ends with Leah. Give me the capital city of that country as well. The answer should be something like Australia and Canara. 

Let's send it and see. Okay, here's the answer and it's correct. Let's mark this as a pass. Now the next question is what is the number that rhymes with the word we use to describe a tall plant. The answer should

Be three. Let's send it and see. Okay, here's the answer and it's Al correct. So this is also a pass. 

The next question is write a Haiku where the second letter of each word when put together spells simple. Let's send it and see. Okay, so here's the answer and this is also correct. So this is a pass. 

The next one is name an English adjective of Latin origin that begins and ends with the same letter, has 11 letters in total, and for which all vowels in the word are ordered alphabetically. The answer could be something like transparent. Let's send it and see. Okay, here's the answer and this is also correct. So let's mark it as a pass. 

The next question is Courtney said that there were 48 people, but Kelly said that Courtney had overstated the number by 20%. If Kelly was right, how many people were there? The answer should be 40. Let's send it.

And check this is the answer and it's correct. Let's mark it as a pass. The next one is I have two apples. Then I buy two more. I bake a pie with two of the apples. After eating half of the pie, how many apples do I have left? The answer should be two. Let's send it and see. Here's the answer and it's correct. So let's give this a pass.

Now the next question is Sally is a girl. She has three brothers. Each of her brothers has the same two sisters. How many sisters does Sally have? The answer should be one. Let's send it and see. Now here's the answer and this is also correct. So let's mark it as a pass as well.

The next one is if a regular hexagon has a short diagonal of 64, what is its long diagonal? The answer should be 73.9. Let's send it and see. Now here's the answer and this is also correct. So let's mark it accordingly.

Question is create an HTML page with a button that explodes confetti when you click it. You can use CSS and JS as well. Let's send it and check. Here's the code and if we run it then this works fine. So this is a pass.

The next question is create a playable synth keyboard using HTML CSS and JS. Let's send it and check. Okay here's the answer and if we run it then this doesn't work. So this is a fail.

The next one is generate the SVG code for a butterfly. Let's send it and see. Okay here's the answer and if we preview it then this doesn't look good. So let's mark it as a fail.

The next one is create a 3D Circle that moves and loops endlessly in 3D space using HTML CSS and JS. Let's send it and check. Okay here's the answer and this works fine. So let's keep this as a pass.

Now the final question is write a game of life in Python that.

Works on the terminal. Let's send it and see. Okay, here's the code. Let's copy it and run it. Okay, it works well. Let's keep it as a pass. 

Now this is the final chart and it was quite good. Coding is not its strong suit, unlike deep seek R1, which is really good at coding. However, it's still good at general questions, although it tends to repeat tokens. The two hardest questions took about three or four generations, and in one instance, it didn't glitch and gave the correct answer. 

So this is good, but it brings nothing above deep seek R1. R1 is still better than this, and it has open weights and is better. So you can use it if you wish to, and it's good. But I would have liked the model to be open source, but it isn't. 

And the APIs are also not available as of now, so there's that. It's a good model, but I can't think of a use for it.

Because it brings nothing new to the table is pretty unusable as of now. And it isn't as great. Their research papers are quite good and have some impressive elements, which is also great to see. Overall, it's pretty cool anyway. 

Note, that there may be slight differences in text and video. Watch video for the original content.

Kimi AI Video