Google DeepMind officially released its Gemma 4 suite on 22 May after a wild week of the online rumors and the leaks. This move follows the sudden appearance of a mystery model on the LMSYS Chatbot Arena on 14 May.
That model used the code name of "significant-otter" and it shocked the testers with its high speed and its very stable outputs. And when the users asked for its name, the AI simply confessed that it was a new project from the team at Google.
Now the company has confirmed the leaks by launching three new versions of the open-weight software for the public use. They built the 2B, 9B, and 27B models to run directly on the consumer laptops and the mobile phones.
But the main goal is to let the developers build tools that work without the expensive cloud server links or the privacy risks. So, the local data stays on the device where it belongs instead of being sent to a distant data center.
The new suite supports over 140 languages and it handles the multimodal tasks like the image processing for the local AI assistants. Yet the speed is the most vital part of this update for the most users in the field. DeepMind says a slow answer is as good as no answer in a real-time setting.
Discussion