Gemma: Google enters the open-weights game
In one sentence Google releases Gemma 2B and 7B, open-weight models derived from Gemini research. For the first time Google competes directly with Llama and Mistral on open ground.
Google has always kept its best models closed (Bard, Gemini). With Gemma it changes course: it publishes two "small" models (2 and 7 billion parameters) with downloadable weights, free, including for commercial use.
They are Gemini's little siblings, trained with the same technology but in sizes that run on a single computer or GPU. They are not better than Llama 2, but it is a political move: Google admits the open community is too important to ignore.
For developers it means one more choice: Llama (Meta), Mistral, and now Gemma. Three competitive open families to build products on without depending on paid APIs.
Companies
Google, Google DeepMind
Tools
Gemma 2B, Gemma 7B
Tags
Sources