Google Releases Gemma 4 12B Multimodal Model
The new 12-billion-parameter open model from DeepMind introduces a unified 'any-to-any' architecture for advanced multimodal tasks.
Google DeepMind has released Gemma 4 12B, a new generation of its open model family. This 12-billion-parameter model is available under a permissive Apache 2.0 license, continuing Google's commitment to providing powerful tools for the open-source AI community.
Unlike many existing vision-language models, Gemma 4 is built on what Google calls a "unified any-to-any" architecture. This design aims to natively handle a wide variety of data modalities for both input and output, moving beyond the common text-and-image limitations of previous systems.
Why 'Any-to-Any' Matters
This architectural approach is significant for developers. It simplifies the process of building complex applications that need to interpret and generate combinations of different data types, such as text, images, and potentially other formats in the future. Instead of chaining together multiple specialized models, developers can use a single, more integrated system, which could enable more fluid and capable AI assistants, creative tools, and analysis engines.
By releasing a model with this advanced multimodal design, Google provides a powerful new foundation for open-source development. Researchers and engineers can now experiment with and build upon this flexible architecture, pushing the boundaries of what's possible with open AI. The model is available now on Hugging Face.
Sources
- Visit
google/gemma-4-12B
Hugging Face
More from Google DeepMind
All Google DeepMind releases →
EmbeddingGemma 2 Expands to Any Modality
Google DeepMind's compact embedding model now spans text, vision, audio, and video in a single vector space.
Google DeepMind's Gemma 4 Goes Multimodal and MoE
The new open-weights family adds a mixture-of-experts design, encoder-free multimodal inputs, and an optional thinking mode.
Google DeepMind Releases TabFM for Tabular Data
A new foundation model brings zero-shot, in-context learning to classification and regression on structured tables.
More in Any-to-Any
All Any-to-Any →
Liquid AI's LFM3-Omni-2B brings any-to-any AI to the edge
The 2-billion-parameter model combines vision, audio and speech-to-speech in a package small enough to run on device.
MiniMax Debuts M3.1 Flash, a Multimodal MoE Model
The new release pairs a mixture-of-experts design with agentic, coding, and video-understanding abilities in a single open model.
IBM's Granite Switch 4.1 Unifies Chat, Logic, Speech
A single endpoint now routes text conversation, decision-making, and speech synthesis across IBM's Apache-licensed Granite line.
0 comments
No comments yet. Be the first to weigh in.