Google AI Updates: April 3, 2026

1. Google Releases Gemma 4: Most Capable Open Model Family Under Apache 2.0

Google. Google DeepMind released Gemma 4 on April 2, 2026, its most capable family of open models to date, available on Google Cloud and designed to run locally on smartphones, laptops, and single-GPU workstations. The family includes four models — Effective 2B, Effective 4B, 26B MoE, and 31B Dense — all released under the commercially permissive Apache 2.0 license, making them freely usable without royalties or usage restrictions. All four support native vision and video processing; the two smallest also handle native audio input. Context windows reach 128K tokens for smaller models and 256K for the larger variants, and the models are reported to outperform Gemma 3 across every major benchmark. Gemma 4 is built from the same research stack as Gemini 3 and is available via Google AI Studio, Vertex AI, Hugging Face, and Kaggle at launch. Source