Musk Commits to Releasing a New Grok Every Two Weeks, But Three Key Modelers All Quit Within a Week

According to Dongcha Beating monitoring, in the past week, xAI saw three key technical core members who directly led the development of the Grok model resign consecutively. Unlike previous departures at the co-founder and executive levels, these three individuals were each the actual heads of the search after training, pre-training, and voice, the three core product lines. Their collective departure will directly impact Grok's ability to iterate on next-generation models.

Tianyi Zhang confirmed his departure from X today. He was the head of the search and factual after-training team and previously served as a machine learning engineering manager at Apple for five years. Under his leadership, the team reduced the fact error rate of Grok's real-time mode by 70% to 80%, helping Grok 4.20 rise to the top of the Search Arena blind test upon release. This search capability has now also been integrated into Tesla's in-car system.

Juntang Zhuang announced his departure on May 9th, having left earlier this year. He was deeply involved in pre-training since Grok 2 and later grew to become the head of the pre-training team, leading the training infrastructure and core formulas for the entire Grok 4 series of models. Prior to joining xAI, he worked at OpenAI for two years, where he was a key contributor to GPT-4o and the inventor of the GPT-4 Turbo 128k long-context algorithm.

David Haxton confirmed his resignation on May 10th, having previously built Grok's Voice Agent from scratch, as well as real-time thinking voice and voice cloning features. Under his leadership, Grok Voice currently ranks first in the Sierra AI τ-Voice Voice Agent benchmark test, surpassing OpenAI and Google.

The timing of this wave of departures is particularly striking. The currently publicly available Grok 4.3 is still only a 0.5T parameter intermediate version. Musk said on April 18th that the real 1T flagship was "about 5 days away from initial training completion," and the next day he clarified that Grok 4.4 = 1T, expected to be released in early May, and Grok 4.5 = 1.5T, aiming for delivery by the end of May. He claimed the model factory was up and running, with a new basic model scheduled for release every two weeks thereafter. However, as we are now well into mid-May, Grok 4.4 has yet to make an appearance, while the actual core team creating these models is rapidly dwindling.

xAI is currently undergoing a systemic talent drain from co-founders to the model's core layer.