CEO Elon Musk announced that xAI is nearing completion of training for Grok 4.6, a new model with 2 trillion parameters expected to outperform the current 4.5 version and potentially compete with Moonshot AI’s Kimi K3. Kimi K3 currently leads the Arena AI rankings with 2.8 trillion parameters. The Grok 4 series is being developed on the Colossus supercomputing cluster, projected to reach a scale of 1 million GPUs. Experts believe the actual performance of Grok 4.6 depends on the data, algorithms, and post-training optimization rather than solely on the number of parameters.
TradingKey – On July 18th, Eastern Time, SpaceX (SPCX) and Tesla (TSLA) CEO Elon Musk revealed on the X social media platform that the training of Grok 4.6, the next-generation large-scale model developed by his AI company xAI, has entered its final stages, with initial training expected to be completed next week. (Note: Musk’s exact words on July 18th were “next week,” but this week is considered based on the July 20th post.)
According to Musk, this new model with 2 trillion parameters “outperforms in every way” the current Grok 4.5 version, which has 1.5 trillion parameters, while its inference speed and token efficiency remain essentially the same as the previous generation. In a response to a user, he stated that this model has the “potential to surpass” Moonshot AI’s newly released Kimi K3.
This statement comes shortly after Chinese AI startup Moonshot AI released the Kimi K3. On July 17th, Moonshot AI officially launched its next-generation large-scale model, the Kimi K3, with a parameter scale of 2.8 trillion, making it the world’s largest open-source model by parameter scale. On the launch day, Mr. Musk commented “impressive” in the comments section of a related review report.
Since its release, the Kimi K3 has performed impressively on third-party review platforms. On the Arena AI frontend programming ranking, the K3 topped the global rankings with a score of 1,679, surpassing Anthropic’s Claude Fable 5 and OpenAI’s GPT-5.6 Sol. Meanwhile, Moonshot AI officially acknowledged that the overall user experience of the K3 still lags significantly behind these two leading closed-source models.
Kimi’s official account responded to Musk’s statement with a post on the social media platform Weibo: “@ElonMusk Welcome to the ‘2 trillion+’ club.” It has been reported that the full weighting of Kimi K3 models will be open to global developers before July 27th. Currently, xAI has not announced an official launch date, full training plan, or technical evaluation metrics for Grok 4.6. Looking back at the research and development (R&D) process, xAI completed training a 1.5 trillion parameter V9-Medium base model in May 2026, which is considered the core foundation of Grok 4.5.
In terms of computational power, models in the Grok 4 series are trained using a super-large AI computing cluster codenamed “Colossus”. In January 2026, xAI announced that Colossus 2 had gone live, equipped at launch with approximately 555,000 GB200 and GB300 series GPUs, equivalent to the computing power of about 1.4 million H100 GPUs. xAI had previously revealed that Colossus 2’s ultimate goal was 2 gigawatts (GW) of capacity, with plans to add over 400 megawatts (MW) of computing power by June 2026, integrating over 220,000 new GB300 chips, aiming for a final deployment of 1 million GPUs.
Industry analysis reveals that while the total number of parameters in a large language model is an important reference for measuring model scalability, the model’s actual overall capability depends on many factors such as the quality of training data, the algorithmic framework, the inference architecture, the supply of computational power, and post-training optimization. Whether xAI can maintain the performance of its predecessor while scaling the model will be key in determining whether Grok 4.6 can pose a substantial challenge to Kimi K3.
&im=FitAndFill=(700,400))
