Yesterday @grass and @inference_net released a 12B video annotation model that is 15x cheaper than closed models and exceeds or matches the performance of GPT-4.1 and Claude 4 Sonnet Pretty nuts especially as it was trained on 1 million video frames from Grass's network. I'm also very excited to see productization of the network. Imagine how good these models can become as Grass/Inference continue to work together and Grass scrapes the entire go forward state of the internet to train models. Imagine🌱