Pasted the same HTML/JS code (330 lines) into Qwen 35B A3B and Gemma 26B A4B. Qwen: tokenized the input to 1609 tokens Gemma: tokenized the input to 4258 tokens. Damn. I've never noticed this before and I haven't seen people mention it. That alone helps explain why Qwen is regarded as better at codi
