Monolith 1.0 - This FR*EE AI Model Claims to Beat GPT 5.6 sol & Claude Mythos

[00:00:00]Guys, welcome back to another new exciting video. Another new open source model, which is Monolith 1.0. And here you see this is the Hugging Face page.

[00:00:07]And this model is from the Basil Labs AI. And they are claiming that their model is beating the GPT 5.6 all and also beating Mythos 5 and also beating Skinny Kid K3 and Opus 4.8 and Gemini 3.5 flash. So, this model is so much powerful and on this IJD GPQA Diamond AIME 25 and also this MMLU Pro, this model is scoring 90% score. And also I have tried this model on their official playground. Here you see for free you can try this model. And the only thing is that per hour you will get 10 chat only. Okay, so each and every hour this number gets refreshed. And the testing that I have done, I I actually found that this model really have the great capability because here you see, let me show you the question that I have tried. So, first I asked ChatGPT that please give me some of the hard reasoning and math and science and long contest analysis question. And here you see, I I tried this question, advanced logical reasoning. And I tried then this question. Which one? Yes. This is a probability based question. And for this here you see the answer is no and the probability remains the answer is 1/3.

[00:01:23]Okay. Now, when I gave this question to this model Monolith 1.0 and here you see that step one, step two, step three.

[00:01:32]Like this way it actually tried and at the end here you see this is the answer that is no. And also here you see it stays the same 1/3. And you see the the ways it actually thought. So, basically this is the problems of this chat. This is interface because at a time you will get some output token, some limited output token.

[00:01:56]That's why here you see that after giving some output, the model got stopped in the midway. But when I asked it that, "Please continue." then it actually uh started from uh where it left. So, it left at uh step three and then it started again from step three.

[00:02:11]And uh then uh step four it got stuck and uh then continuing on step four. So, basically this is the limit of their chat interface, not the model itself. Uh because if you see uh this monolith uh this uh context window is around 1 million. And also 128 experts and 1.6 trillion total parameter and 49 billion active parameter. Okay, and the knowledge cutoff is still December 25.

[00:02:38]Okay. So, the model is great. Okay. Uh uh means if you are trying this model inside this chat interface, then you will face this kind of issue that after giving some output the model will be stuck. So, you have to just uh then write this continue. And one another uh limitation in their chat interface is that uh they are supporting uh 1,024 tokens uh for uh means per chat per chat. Okay. Uh because of their uh maybe infrastructure limitation. But otherwise overall uh what I have uh found that the model really have the great capability. Now, another question I I actually gave it.

[00:03:13]Uh this is the question. Where is that man? Uh Yes. This one. Advanced logical uh reasoning question. But they are uh They actually the model itself actually thought a lot and uh we reached the number of tokens inside their chat interface. So, that's why now we we did did not get the uh ultimate answer. But here you see the thought process actually was very good. From lots of angle it is uh thinking. Okay.

[00:03:40]So, the claiming that uh they are doing that they are uh beating all of these popular GPT-5.6B and Mixtral 8x7B on these reasoning areas, it may be true because we all know that Mixtral 8x7B and GBD 5.6 all they are actually for the cyber security purpose coding forecast. Okay, so more more agentic purpose long running step-by-step process completion purpose. They have made this model for that purpose only.

[00:04:06]Okay, so and and this labs AI model they have made this only for this reasoning math science and long context analysis purpose. The purpose is different for all of this. So that's why the benchmark that they are claiming it may be 100% true and this model also getting GBD 5.4 or pass 4.6 73.1 Pro K2.6 and and the and then the difference between this percentage actually huge 99.4% and it is 40% in case of GBD 5.4 and if you see that for GBD 5.6 all it is around 64%.

[00:04:43]So the difference is huge and yes, this is the actual thing. So you also please try it and let me know your So yes, you also please try it on their platform vessel.org. I I will give this link in description. You can go there do the sign up. Okay, after sign up you will actually get this this 10 message limitation. Okay, another thing they have also their max version. So if you just click on this you will enable this max otherwise it will use the normal chat process. If you enable max then it will think a lot.

[00:05:16]Okay, and at that time it will give the response I means you will get the give you will get you will get the response after some time means there will be some delay. Okay.

[00:05:29]Okay, so yes, these are the information that I wanted to share with you guys. If you found this video helpful, don't forget to subscribe this channel. Don't forget to like this video also to get such information daily all of the latest AI related news. See you guys in the next video. Thanks for watching.

[00:05:42]Bye-bye. Take care.