Thanks for the effort and honesty.
#4
by
phil111
- opened
Edit: My original comment was deleted because I realized this model uses the QwQ-32 base, and that's the only reason it has more broad knowledge than Qwen3 34b.
It's interesting watching the progression of Alibaba's Qwen series. Qwen2 72b had nearly as much broad popular knowledge as Llama 3.1 70b (e.g. a SimpleQA score >20), but Qwen2.5 72b's broad knowledge plummeted (e.g. a SimpleQA score of 10). And now even the massive Qwen3 252b has less general knowledge than Qwen2.5 72b despite being far larger, and Qwen3 34b only has the broad knowledge of 1b models, scoring lower than Gemma 2 2b on my broad knowledge test.
phil111
changed discussion status to
closed