Google DeepMind releases Gemma 4 12B: encoder-free multimodal model runs on 16GB RAM
Google DeepMind has released Gemma 4 12B, a 12-billion parameter multimodal model that runs locally on laptops with 16GB of RAM. The model eliminates separate vision and audio encoders, processing raw inputs directly through its language model backbone under an Apache 2.0 license.
Google DeepMind releases Gemma 4 12B: encoder-free multimodal model runs on 16GB RAM
Google DeepMind has released Gemma 4 12B, a 12-billion parameter multimodal model designed to run locally on consumer laptops with 16GB of RAM. The model eliminates traditional multimodal encoders, processing vision and audio inputs directly through its language model backbone.
Technical architecture
Gemma 4 12B differs from conventional multimodal models by removing separate encoder modules:
- Vision processing: Replaces the vision encoder with a single matrix multiplication, positional embedding, and normalization layers, allowing the LLM backbone to handle visual processing directly
- Audio processing: Projects raw audio signals into the same dimensional space as text tokens without any encoder
- Memory footprint: Requires 16GB of VRAM or unified memory for local inference
According to Google DeepMind, this architecture reduces latency and memory usage compared to encoder-based approaches.
Performance and positioning
Google DeepMind claims Gemma 4 12B delivers benchmark performance approaching its larger 26B Mixture of Experts model at less than half the memory footprint. Specific benchmark scores were not disclosed. The model sits between the company's edge-focused E4B and the 26B MoE in terms of capability and size.
Gemma 4 12B is the first mid-sized model in the Gemma family to support native audio inputs alongside text and vision.
Availability and ecosystem support
The model is released under an Apache 2.0 license and available now through:
- Direct download from Hugging Face and Kaggle
- Inference tools: LM Studio, Ollama, Hugging Face Transformers, llama.cpp, MLX, SGLang, and vLLM
- Fine-tuning: Unsloth support
- Deployment: Google Cloud via Gemini Enterprise Agent Platform Model Garden, Cloud Run, and GKE
Google also released an official Skills Repository library for agent development with Gemma models.
The company states the Gemma 4 model family has reached 150 million downloads. The model includes Multi-Token Prediction drafters for latency reduction.
What this means
Gemma 4 12B represents a shift toward unified multimodal architectures that eliminate specialized encoder modules. By processing raw audio and simplified vision embeddings directly in the LLM backbone, Google is betting on architectural simplicity over modular design. The 16GB memory requirement makes this genuinely laptop-deployable for developers, though actual performance relative to encoder-based alternatives remains to be independently verified. The Apache 2.0 license and broad tooling support position this as a practical option for local multimodal inference.
Related Articles
Mistral AI Releases Shieldstral-1.0-3B, a 3B-Parameter Policy-Adaptive Safety Classifier
Mistral AI has released Shieldstral-1.0-3B, a compact open-weight safety classifier that evaluates text and images against natural-language policies specified at inference time. The 3B model runs on a single GPU and reports F1 scores competitive with or exceeding larger moderation models like LlamaGuard-4-12B and GPT-OSS-Safeguard-20B on multiple benchmarks.
Mistral Releases Shieldstral, a 3B Open-Weights Safety Classifier That Matches Models 7x Its Size
Mistral has released Shieldstral, a 3B open-weights safety classifier that reframes content moderation as a policy-adaptive question-answering task. The model claims to match or outperform guard models up to 7x its size on text safety and multimodal benchmarks, and runs on a single 16GB GPU.
OpenAI Halts Parts of Astra Model Development After It Hit 'Critical' Cybersecurity Threshold
OpenAI disclosed that its in-development Astra model showed cyberattack capabilities strong enough that it cannot rule out a 'Critical' risk classification. The company has paused related internal activity and added security controls under its Preparedness Framework.
Mistral's 3B-Parameter Shieldstral Matches 20B Safety Model on Text Benchmarks
Mistral's new Shieldstral, a 3-billion-parameter open-weight safety classifier, posts an 84.9% F1 score on text benchmarks—tying OpenAI's GPT-OSS-Safeguard-20B, a model roughly seven times larger. The model lets operators define safety rules at runtime using plain-language yes/no questions instead of fixed taxonomies.
Comments
Loading...