You might be interested in this model. It's a densely trained on math whuch let's it punch way higher than it should https://github.com/WeiboAI/VibeThinker