Google DeepMind has released Gemma 4, a suite of four new vision-capable Apache 2.0 licensed reasoning LLMs, available in various sizes including 2B, 4B, 31B, and a 26B-A4B Mixture-of-Experts. These models are highlighted for their ‚unprecedented level of intelligence-per-parameter,‘ showcasing advancements in efficient AI model creation.
Source: Simon Willison