Gemma-3-12B

Gemma 3 12B is a well-balanced mid-sized multimodal language model developed by Google DeepMind, designed to tackle narrow, specialized professional tasks. With 12 billion parameters, the model combines high performance with computational efficiency and supports a wide range of capabilities—from text analysis to image processing. Gemma 3 12B converts visual data into tokens, enabling deep understanding of images. The "Pan&Scan" technology allows adaptive processing of images with any aspect ratio, preserving detail when scaling up to a resolution of 896×896.

Another key feature is the expanded context window of up to 128K tokens. This enables the model to process lengthy legal documents and scientific articles in a single request without losing context. Multilingual support covers more than 140 languages, including Russian, while the enhanced tokenizer from Gemini 2.0 ensures high-quality translation, text generation, and cross-lingual analysis. Additionally, developer-supported quantization makes it possible to run the model even on consumer-grade GPUs with minimal loss in quality.

As a result, Gemma 3 12B is a versatile tool for data analysis, document processing, and information extraction from visual sources—with the ability to run locally and scalable integration into modern AI infrastructures.

Announce Date: 12.03.2025
Parameters: 12B
Context: 131K
Attention Type: Sliding Window Attention
VRAM requirements: 16.4 GB using 4 bits quantization
Developer: Google DeepMind
Transformers Version: 4.50.0.dev0
Ollama Version: 0.6
License: gemma

Use our pre-built public endpoints to test inference and explore Gemma-3-12B capabilities.

Rent your own physically dedicated instance with hourly or long-term monthly billing.

We recommend deploying private instances in the following scenarios:

• maximize endpoint performance,
• enable full context for long sequences,
• ensure top-tier security for data processing in an isolated, dedicated environment,
• use custom weights, such as fine-tuned models or LoRA adapters.

Contact our dedicated neural networks support team at nn@immers.cloud or send your request to the sales department at sale@immers.cloud.