ChainThink reports that on July 18, after the release of Kimi K3, Musk disclosed progress on SpaceXAI’s next large model with 2 trillion parameters, stating that initial training will be completed next week.
Musk stated that the new model outperforms Grok 4.5 with 1.5 trillion parameters in all aspects and may surpass Kimi. He expects that, while enhancing capabilities, the model will achieve generation speed and token efficiency close to that of Grok 4.5.
Artificial Analysis tests show that Kimi K3 and Grok 4.5 have intelligence scores of 57 and 54, respectively; the cost per task for each is approximately $0.94 and $0.31.
