Gemma is Google’s family of open-weight models, the open counterpart to its API-and-product Gemini line. Google’s launch announcement describes Gemma as “a family of lightweight, state-of-the-art open models built from the same research and technology used to create the Gemini models,” designed so developers can run them “wherever users need them - from cloud servers to laptops and even phones.” Where Gemini is delivered through Google’s APIs and products, Gemma weights are released for download, fine-tuning, and self-hosting.
The line began in February 2024 with two sizes, Gemma 2B and Gemma 7B, each in pre-trained and instruction-tuned variants, which Google said achieved “best-in-class performance for their sizes compared to other open models.” Subsequent generations followed, and Google’s Gemma site presents the family as “byte for byte, the most capable open models.” Over time Gemma has expanded beyond the core text models into a set of specialized open variants - for example models aimed at medical text and image understanding, on-device embeddings, content-safety classification, and privacy-preserving training. Version numbers and the exact current lineup advance quickly, so the specific models named here reflect Google’s announcements as of the verification date; Google’s Gemma page is the live reference for the current catalog and sizes, and parameter counts are not frozen.
Distribution is open weights: the models are downloadable and self-hostable, which is the defining contrast with the Gemini family.
Why business readers should care: Gemma gives organizations a Google-backed open-weight option they can run on their own infrastructure, useful where data control, cost, offline operation, or on-device deployment rule out a hosted API - while sharing lineage with the Gemini models many teams already use.