Gemma 4 31B (IT)

Model ID: vllm/release/gemma-4-31b-it Google Productive

EU AI Act

Attribute	Value
Classification	GPAI (General-Purpose AI Model)
Risk class	Limited Risk (Article 52)
Transparency obligation	Yes
noris role	Deployer

As a GPAI model, this model falls within the scope of EU AI Act transparency obligations. noris acts as the deployer and ensures the required information is made available pursuant to Article 53.

Technical Specifications

Property	Value
Model name	Gemma 4 31B (IT)
Provider	Google
Model ID	vllm/release/gemma-4-31b-it
Parameters	~31B
Architecture	Dense Transformer (Gemma 4)
Context length	262,144 tokens
Modalities	Text + image (multimodal)
Quantization	NVFP4
License	Gemma Terms
Release	June 2026

Recommended Use Cases

• General chat
• Classification
• Text analysis
• Image description
• RAG generation

Known Limitations

Caution

• Knowledge cutoff early 2026: the model has no knowledge of events after its training period.
• Gemma Terms (not Apache): usage is subject to Google’s proprietary Gemma terms, which may include restrictions.
• NVFP4 can introduce slight quality loss: quantization may cause minimal accuracy degradation in individual cases.

Risk Assessment (EU AI Act)

Risk Factor	Category	Level
Open source	Misuse potential	Medium
Multimodal / image input	Deepfake potential	Medium
Training data	Copyright	Medium
Environmental impact	Model size	Low: NVFP4 efficient

Training Data & Sustainability

Training data: not fully publicly documented; web crawl, licensed content.

Sustainability:

Metric	Value
CO₂ (estimated)	~10 to 50 tCO₂e
Optimization	NVFP4 quantization for efficient inference

Original Sources

• HuggingFace: https://huggingface.co/google/gemma-4-31B-it
• License: Gemma Terms

Model card created on July 2, 2026. This model card serves EU AI Act transparency purposes (Regulation (EU) 2024/1689). The information is based on publicly available data from the model provider.