DiffusionGemma Shows How Existing Language Models Can Be Repurposed for Faster Text Generation
Google DeepMind adapted Gemma 4 into DiffusionGemma at a fraction of the training cost, trading benchmark quality for faster parallel text generation.
Latest News and Analysis in Gemma 4
Google DeepMind adapted Gemma 4 into DiffusionGemma at a fraction of the training cost, trading benchmark quality for faster parallel text generation.
Google says Gemma 4 has passed 150 million downloads and showcased builders using the open models for on-device AI apps.
Google DeepMind launched Gemma 4, a new family of open-weights AI models optimized for agentic workflows and coding, now under a permissive Apache 2.0 license.