AnalysisAI ModelsJune 25, 2026
Gemma4 QAT Balanced models boost speed 35-53%

Two new fine-tuned Gemma4 models (26B-A4B and 31B-QAT) with Multi-Token Prediction achieve 35% and 53% speed boosts. The author's HuggingFace account has nearly 20 million downloads and 5,000 Discord members.