The catch is that it only has ~5 billion active params so should perform worse than the top Deepseek and Qwen models, which have around 20-30 billion, unless Op

The catch is that it only has ~5 billion active params so should perform worse than the top Deepseek and Qwen models, which have around 20-30 billion, unless OpenAI pulled off a miracle.