Keep pulling the thread on Demis Hassabis.
The 12B Gemma 4 model features a unified, encoder-free architecture that ingests raw audio and image patches.
Gemma 4 models integrate a "thinking mode" that enables them to generate reasoning traces before responding.
Gemma 4 demonstrates a significant performance improvement across STEM, multimodal, and long-context benchmarks.
Gemma 4's performance on human-rated tasks is competitive with larger, frontier open models.
Gemma 4 is a new generation of open-weight, natively multimodal language models in the Gemma model family.
The Gemma 4 model suite includes both dense and Mixture-of-Experts (MoE) architectures.
The Gemma 4 model suite features models with parameter sizes ranging from 2.3 billion to 31 billion.
Gemma 4 has improved inference speed, memory efficiency, and compute efficiency due to its design choices.
Gemma 4 has improved long-context abilities due to its design choices.