11h ago
Google releases Gemma 4 12B, an Apache 2.0-licensed encoder-free multimodal model that outperforms the larger Gemma 3 27B
AI Judge changed title after evaluation, original title: "Google launches Gemma 4 12B, a multimodal open model that outperforms the larger Gemma 3 27B"
The model runs locally on laptops within 16GB VRAM.