RESEARCH · GLOBAL
Google DeepMind releases DiffusionGemma: 4x faster text generation
DeepMind published DiffusionGemma, a new approach to text generation using diffusion models that achieves 4x speedup over standard autoregressive decoding, reducing inference latency and cost.
WHY IT MATTERS
Inference efficiency breakthrough. BFSI models serving customer-facing queries can reduce latency and per-token costs; enables real-time agent decision-making in trading, fraud, and customer service at scale.
Source: Google DeepMind · 2026-06-10