Elon Musk announced on September 14 that Grok 4.8, an AI model with 2.5 trillion parameters, is wrapping up its initial training run and entering the reinforcement learning phase. This version follows Grok 4.6, released a month earlier with 1.5 trillion parameters, illustrating xAI’s rapid release cadence. The company is building models on a 3 trillion-parameter architecture, with potential variants reaching up to 10 trillion parameters. xAI relies on Colossus 2, described as the world’s first gigawatt-scale training cluster, and leverages proprietary data from SpaceX operations and Cursor engineering workflows. This monthly release strategy contrasts with the quarterly or semi-annual cadence of competitors.
Source: Read the original article
