Elon Musk announced the completion of training for a new language model from xAI. According to him, the neural network, boasting 2 trillion parameters, will finish its first stage of preparation as early as next week. The founder of SpaceX and Tesla emphasized that the new product will surpass the current version of Grok 4.5 "in every way".
Architecture Battle: Efficiency vs. Scale
Musk cited the combination of quality and speed as the key advantage of the new development. He noted that the model is capable of outperforming the Chinese counterpart Kimi in terms of performance quality, while maintaining the high generation speed and computational resource efficiency characteristic of Grok 4.5.
The trigger for widespread discussion was a comparison conducted by researcher Min Choi. In his analysis, he pointed out that the current version of Grok 4.5 uses about 1.5 trillion parameters, whereas the Kimi K3 model from Moonshot AI counts 2.8 trillion. At the same time, Choi stated that running tasks on Kimi costs approximately three times more. The researcher called this result evidence of the high efficiency of the xAI architecture, allowing for better results with lower resource costs.
Competition Context and New Records
Previously, the Chinese company Moonshot AI presented its new artificial intelligence model, Kimi K3. The developers claimed it became the largest open-source AI model in the world, sparking a wave of interest in the capabilities of Chinese neural networks.
At the same time, Elon Musk reported new achievements for the current product line. The Grok 4.5 language model took first place in the Long-Horizon Terminal-Bench ranking as of July 2026. This success confirms xAI's leadership in creating efficient and powerful language models capable of competing with the world's largest players.
