Gemma 4 MoE

Google released DiffusionGemma, an open-weight model that generates text in parallel 256-token blocks, achieving 1,500 tokens/second on a single H100, fine-tuned from Gemma 4 MoE (3.8B active / 25.2B …

A developer built a search engine over 50k saved tweets using hybrid retrieval and reranking, then added a single endpoint to turn it into an answer engine. By wiring Gemma 4 MoE to the retrieval pipe…