Google DeepMind is simulating entire worlds using AI

SynthID, our imperceptible watermark for AI-generated content, is expanding to more partners.

Gemma 4 is Google DeepMind’s latest open multimodal model family, aimed at both on device and broader deployment, with variants including E2B, E4B, a 26B Mixture of Experts model, and a 31B dense model Google claims it can handle text, images, audio, video, long context reasoning up to 256K tokens, multilingual inputs across 140 languages, OCR, chart/document understanding, code generation, tool use, and agentic workflows, while its benchmark highlights include strong results on MMLU Pro, GPQA D