Google releases Gemma 4 12B with on-device multimodal agentic capabilities
Google released Gemma 4 12B, a multimodal model optimized for on-device deployment with agentic workflow support, using an encoder-free architecture to reduce resource overhead and latency. The model targets enterprise and consumer applications requiring local inference without cloud round-trips.
The release expands Google's portfolio in the edge-AI market competing with OpenAI's local ChatGPT, Anthropic's Claude Haiku, and open models like Mistral 7B, emphasizing on-device capability as a privacy and sovereignty differentiator.