At the SpaceX 2026 Q2 earnings call held today (August 5), covering the quarter ended June 30, Musk announced that the Grok 4.6 model will be released next week.
At the earnings call, Musk said that the Grok series of AI models is progressing rapidly. The Grok 4.6 model planned for release next week will have a total of 1.5 trillion parameters, with a focus on improving supervised fine-tuning (SFT) and reinforcement learning (RL).
Note: SFT refers to a technique that adjusts model parameters using high-quality labeled data after the pretraining of a large model, enabling it to precisely adapt to specific tasks and human interaction patterns.
Pretraining is like having a large model read every book in a library (learning general language and common sense), while SFT is like a teacher guiding it through exercises with “standard answers,” teaching it to respond according to rules and formats.
Musk also previewed the release of the Grok 4.7 model in the next 3~4 weeks. It will have 2.1 trillion parameters and be superior in every respect.