Google introduces Gemma 3 270M, a new compact AI model with 270 million parameters that companies can fine-tune for specific tasks. The model promises significantly lower operating costs than larger alternatives.
Gemma 3 270M is the latest addition to Google’s open source AI model series. With 270 million parameters – 170 million for embedding and 100 million for transformer blocks – the model has been specially developed for task-specific fine-tuning. The large vocabulary of 256,000 tokens ensures that the model can also handle specific and rare terms well.
The model stands out for its energy efficiency. Tests on a Pixel 9 Pro SoC showed that the INT4-quantized version consumes only 0.75 percent battery power for 25 conversations, “making it our most power-efficient Gemma model,” according to Google.
The small size of Gemma 3 270M enables rapid fine-tuning experiments. Where other models take days, Google now claims that companies can find the perfect configuration for their use within hours. In addition, the model can run entirely on-device, which means that sensitive information never has to leave the cloud.
Widely applicable for specific tasks
Gemma 3 270M focuses on high-volume, well-defined tasks such as sentiment analysis, entity extraction, query routing, and processing unstructured to structured text. For companies looking to save every millisecond and micro-cent, this model offers dramatic cost savings. A fine-tuned 270M model can run on lightweight, low-cost infrastructure or directly on devices.
The model is available immediately via platforms such as Hugging Face, Ollama, Kaggle, LM Studio, and Docker. Both pre-trained and instruction-tuned versions have been released. For deployment, organizations can choose from local implementation to Google Cloud Run.
Google also offers production-ready quantization-aware trained (QAT) checkpoints, which allow models to run at INT4 precision with minimal performance loss. This approach is essential for deployment on devices with limited resources.
For companies looking to build a fleet of specialized task models, Gemma 3 270M offers a cost-effective foundation. Each model can become an expert in its own task without breaking the budget.