xAI Completes Training on Grok V9 Medium, Targets Mid-June Public Release

xAI has reached a significant milestone with the completion of training for Grok V9 Medium, its advanced foundational AI model. Announced by Elon Musk on May 25, 2026, this new iteration features an impressive 1.5 trillion parameters, making it three times larger than the Grok v8 small model currently deployed in production, which operates on 0.5 trillion parameters. The initial evaluation results for Grok V9 Medium have been described as positive, indicating strong performance. Following the training completion, Grok V9 Medium is now undergoing supervised fine-tuning, a critical phase to refine its capabilities and align its outputs with desired behaviors. Reinforcement learning is slated to begin in the immediate future, further enhancing the model's intelligence and adaptability. The public can anticipate the release of Grok V9 Medium approximately two to three weeks from Musk's announcement, placing its expected launch sometime in mid-June 2026. In a move consistent with its previous strategies, xAI also intends to open-source the Grok v8 small model, the predecessor to V9 Medium, by the close of 2026. This follows a pattern established with earlier Grok versions, where previous-generation models are made publicly available once newer iterations stabilize. For instance, Grok 1 weights were released under an Apache-2.0 license in March 2024, and Grok 2.5 was published on Hugging Face in August 2025. This open-source approach aims to foster broader innovation and research within the AI community. Notably, Grok V9 Medium was also trained on Cursor data, integrating real developer workflows from a popular AI code editor, suggesting a focus on coding prowess.