AI News For Match 12
Google releases Gemma 3, the most capable small AI model in the world. And Gemini Robotics models that will power a new wave of robots. Plus, FreePik gets image to video for Google’s VEO 2 it’s all about Google in today’s AI news. There were two big announcements from Google today.
First it was Gemma 3, a series of lightweight open source AI models that leverage the same technology as Gemini 2. There are four different sizes available from 1 billion to 27 billion parameters and and they all have a 128K context window and multimodal understanding.
Google claims that Gemma 3 is “the most capable model you can run on a single GPU or TPU” and that seems like a claim that they can back up as the 128k context alone is really outstanding. The second announcement was for Gemini Robotics. This is an Advanced Vision Language Action model or VLA that is built on Gemini 2 and it adds physical actions as a new output modality.
This means that it can take an input prompt and combine it with real world image data to produce physical actions even with tasks that it has never been trained on. Google are also partnering with Apptronic, an American robotics company to test the real world application of these models.
Finally, Freepik has scored another big first. Today they announced that image to video for VEO 2 is now available on their AI suite. This is the second exclusive release that they’ve had with Google VEO and these kinds of partnerships are really helping them stand out as one of the leading reading AI, image and video generation tools.