Unbelievable! Gemma 4 —— 31B (Dense) and Gemma 4-27B A4B (MoE, ~26B total parameters, active 4B) ranked #27 and #40 respectively on the @arena (LMSYS Arena) leaderboard. Dense models outperform MoE models! Gemma 4-31B scored higher than the following models: ☑️ GPT-5.1 ☑️ GPT-5.2 ☑️ GPT-5.2-high ☑️ Qwen3.5-397B-A17B (a model with a much larger parameter scale)