History
Gemma 3 targets models for a single accelerator
Google presented Gemma 3 as a model family that could also run on a single GPU or TPU.

On 12 March 2025, Google presented Gemma 3. The company emphasised models that could run on a single GPU or TPU. These are computing chips used for tasks including running models. The family offered models in different sizes. Google explained the release.

For a business considering its own installation, the required equipment is a particularly useful starting point. Alongside answer quality, the required equipment became central. A single accelerator is a technical description; it does not fully describe a managed service.

A small or medium-sized business considering its own installation can start with a proposal for a limited task. Consider labelling public sample texts. Ask which model size will be used and how many requests need to work at once. Then have the same trial run on the proposed equipment. Measure waiting time and check the labels against a list prepared beforehand. Also ask who carries out updates and resolves faults. This connects an announcement about smaller models to the concrete question of whether a suitable installation can be maintained for your business.