- Published on
Musk teases Grok 4.6 model arriving next week, improving AI supervised fine-tuning and reinforcement learning
- Authors
- Name
- DeepHorus
Elon Musk revealed during SpaceX's Q2 2026 earnings call that his artificial intelligence company xAI will release the Grok 4.6 model next week, and previewed future versions. The news was announced during the conference call held on August 5 local time, with Musk also emphasizing the rapid iteration of the Grok series models.
Musk stated during the meeting that the Grok series AI models are progressing rapidly. The upcoming Grok 4.6 has a total of 1.5 trillion parameters, with key improvements in supervised fine-tuning (SFT) and reinforcement learning (RL) techniques. He also revealed that the Grok 4.7 model will be released within the next 3 to 4 weeks, with parameter count reaching 2.1 trillion, and performance expected to be superior across the board.
Supervised fine-tuning (SFT) is a technique that adjusts model parameters using high-quality labeled data after pre-training, aiming to make the model precisely adapt to specific tasks and human interaction habits. In simple terms, pre-training is like having a large model read massive amounts of books to grasp general language knowledge, while SFT is like a teacher guiding based on standard answers, allowing the model to learn to respond in a standardized format.
Musk did not provide a specific release date for Grok 4.6, but confirmed it will launch next week. Regarding Grok 4.7, he only said its performance will be comprehensively improved, without disclosing more technical details. This release comes at a time of intense competition in the AI field, and the rapid iteration of the Grok series may have a certain impact on the existing market landscape.
From a technical perspective, the Grok series continues to strengthen SFT and RL, reflecting xAI's emphasis on model instruction-following and reasoning capabilities. Industry analysts believe that the combination of larger parameter counts and optimized training methods is expected to improve the model's performance on complex tasks. However, the actual results still await user testing feedback.
It is not common for Musk to discuss AI models during an earnings call, as SpaceX's main business is aerospace, but this move also reflects his strategic investment in AI technology. In the coming weeks, with the successive release of Grok 4.6 and 4.7, the outside world will have the opportunity to examine xAI's technological achievements.
Currently, xAI has not disclosed the latest user scale or commercial application data for the Grok series. In the fiercely competitive AI market, whether Grok can stand out with rapid iteration remains to be seen.
