The Two Models That Never Met. Both Measured at the Same Depth.
Gemma4 and Qwen3 were trained by different organizations on different data with different architectures. Their internal representations are 99.2% similar at matched depth. Neither model knew the other existed.
Read More