Google releases Gemma 4 12B AI model that runs on 16 GB

Google has launched the Gemma 4 12B, a compact multimodal AI model that operates on standard devices with only 16GB of memory. Key features include:

• Unified processing of text, images, video, and audio without separate encoders.
• Performance comparable to larger models such as the 26B variant, enhancing cognitive reasoning for complex tasks.
• Open-source release under the Apache 2.0 license, promoting community support.
• Special token prediction selectors that reduce interaction latency.

The model is available for deployment on platforms like Ollama, HuggingFace, and through the Unsloth framework.