Google has released Gemma 4, a new suite of open-source models available under an Apache 2.0 license, addressing previous enterprise concerns. The models come in four variants, including efficient E-series models with native audio capabilities and larger 26B and 31B parameter models. A notable feature is the 26B MoE (Mixture-of-Experts) variant, which achieves impressive inference speeds by activating only a fraction of its parameters per token, making it competitive with smaller models in terms of compute while retaining higher reasoning quality. AI
IMPACT New open-source models with flexible licensing and efficient MoE architectures could accelerate local AI deployment and experimentation.
RANK_REASON New model release from a frontier lab (Google) with system card details. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →