Transcription
Hey everyone, welcome back to another new exciting video.
Mistral has released another two small but powerful models, and their names are Magestral Small 1.2 and Magistral Medium 1.2. These both are open source, and here you see that you will find this model in the name of Magestral Small 259. Here you see this is the GitHub repository, and they have written Magestral Small 1.2.
Basically, they have built this model as an upgrade version of Mistral Small 1.1. Previously, that 1.1 had some capabilities, but they have now added some more extra capabilities, and the model now is 1.2. The newest added feature is this multimodality. Here you see, now equipped with a vision encoder, means you can use both text and images seamlessly.
Next one is this performance boost: 15% improvement on math and coding benchmarks, and also smarter tool use, better tool usage with web search, code interpreter, and image generation. Now, this is a great advantage for the uh coder, or you can say that if you are interested in AI tools or generative AI, then you can use this ML Small 1.2 too because it is open-source, and you will get this tool usage capability because most of the times when we are building one agent, their tool usage is very important, and this small model is giving that capability. So you should definitely use this model.
And here you see better tone and persona: responses are clear, more natural, and better formatted for you. And if you go to this artificial analysis index, and also here you see that reasoning and knowledge, scientific reasoning like codebench, coding, and reasoning and knowledge. So they have compared this model with the other models, and what they have seen that uh this small model can beat GPT OSS. You know that GPT OSS is the 20 billion model, but uh this Magistral Small 1.2 can beat that model also. Okay. And also here you see that uh here also this Magistral Small model is beating that model. So this is the scientific reasoning, and this is the reasoning and knowledge.
And if I show you another interesting thing, here you see they have shared these benchmarks. So they have compared this DeepSeek R1 with this Magestral Medium and also Magestral Medium 256 and Co3 to 35 billion model. And what they have found that Magestral Medium model has the capability to give the same kind of performance like the DeepSeek R1 and 3 to 35 billion. Mean you can imagine that the Magestral Medium 1.2 is such a small model, but it can give the same kind of capability of 3 to 35 billion. 235 billion is huge, but Magestral one Medium is uh is much smaller and has the less size than 3, but is giving the same kind of capability, and also is free and open-source and available in Hugging Face. Okay.
Now, how to use this model? One option is to download this model from Hugging Face locally and use it. Another option is to use this model through LM Studio. So if you install this LM Studio and on the left-hand side, if you click on this search icon, then you will find this model Magistral Small 25509. And if you want to use it through API, so here you see that I have tested this model through API in my Postman. So this is the actually call. So if you copy this call from the description and paste it in your Postman in this section, and after that, if you click on this send, you will get the output.
Now, let me tell you that how to use this call. Here you see if you copy this call from the description. So in this token option, here you see that uh this is the Bearer token, and here you have to give your Mistral API key. Just these things you have to change because this is my API key, and after this video, I will delete this API key. So if you want to create your own API key, it is completely free. Just go to this website. I have given this link in the description, and here on the right-hand side, click on this create new API key, and it will open a box, and here you can give any name here, and after that, uh just um create your API key. Now copy that API key and paste that API key here in this section here. And after that, copy this whole card from here and uh go to your Postman and there just uh just paste it like here in this way, and here if you just uh click on this send, you will find that you will get the answer. Okay.
So let me just run it. Click on this send. And here you see that what I have asked. I have asked that uh what is the best uh let me show you what is the best French cheese. This is this was my question, and after that, here you see that I have got this output. This is the text. Okay. The question is about best French cheese. France is famous for its wide variety of cheese, blah blah blah. A lot of description it has given. So you can use this response in API also, and uh if you are in development, they will and then you will understand this in a better way. Okay.
So these are the procedures, and uh I hope that this detailed explanation and how to use it and all of the things are helpful for you. So if you found this video helpful, don't forget to like this video. Don't forget to subscribe to this channel, and see you guys in the next video. Thanks for watching. Bye-bye. Take care, and please watch the other videos also.