Qwen tokenizes same code to 1,609 tokens vs Gemma's 4,258
A user pasted the same 330-line HTML/JS snippet into Qwen 35B A3B (1,609 tokens) and Gemma 26B A4B (4,258 tokens). The single-sample test suggests tokenizer efficiency helps explain Qwen's stronger coding reputation.
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- AI detectors face criticism over reliability and impact on trust
- Student uses $3 chip to run Claude Code for automated betting
- MiniMax H3 CLIP swap cuts VRAM from 15.7 GB to 4.5 GB
- Artist's AI-generated 'Found [You?]' footage project blends video and music
- Anthropic's Haiku 4.5 nears 12 months without an update