Summary

Google DeepMind's Gemma 4 introduces groundbreaking open-source AI models with unprecedented performance at small scales, ranging from compact on-device models to larger 26B and 31B versions. The models, released under an Apache 2.0 license, are designed to be easily accessible to developers and feature advanced capabilities like multimodal reasoning, autonomous workflows, and efficient mixture-of-experts architecture. The 31B dense model, in particular, stands out by ranking third on the global AI leaderboard while outperforming much larger models, demonstrating significant technological advancement in compact AI systems.