[AINews] Gemma 4: The best small Multimodal Open Models, dramatically better than Gemma 3 in every way

Google DeepMind released Gemma 4, a family of open-weight multimodal models with Apache 2.0 licensing — a significant upgrade from Gemma 3. The 31B dense variant benchmarks alongside much larger MoE models like Kimi K2.5 (744B) and GLM-5 (1T), with native video, image, and audio input. Smaller E2B/E4B variants support audio input for on-device deployment, raising speculation about an Apple/Siri integration.