Musk Claims Colossus 2 is Training 7 Models Simultaneously, with a Maximum of 10 Trillion Parameters

According to monitoring by 1M AI News, Elon Musk, founder of SpaceX and xAI, posted on X that the SpaceXAI supercomputing cluster Colossus 2 is currently training 7 models simultaneously: 1. Imagine V2, the next generation of image and video generation models; 2. Two variants of a 1 trillion parameter model; 3. Two variants of a 1.5 trillion parameter model; 4. A 6 trillion parameter model; 5. A 10 trillion parameter model. He added, “Some catching up to do.” Previously, several media outlets reported that xAI’s next-generation flagship model Grok has a parameter scale of about 6 trillion, which aligns with the 6T listed. The 10 trillion parameter model has not been publicly reported before. In his post, Musk referred to the merged entity as “SpaceXAI.” The two companies completed their merger in February of this year, with a post-merger valuation of approximately $1.25 trillion. Colossus 2 was launched on January 17 and is the world’s first gigawatt-level AI training cluster, with plans to upgrade to 1.5GW this month.