xAI's Grok 5 -- rumored to be a roughly 6-trillion-parameter, mixture-of-experts model, about double the scale of Grok 4 -- remains in training as of mid-June 2026, having slipped past both its Q1 and Q2 target windows. Reporting now points to full API access landing in the third quarter, and prediction markets put the odds of a public release before June 30 at around 7%.
The delay carries real competitive cost. Every quarter the flagship stays in the lab, rivals Anthropic, OpenAI, and Google convert more enterprise customers to their models as defaults -- the stickiest position in AI. Grok 5 will reportedly be natively multimodal across text, images, audio, and real-time video, but capability on paper doesn't win deals that have already been signed.
“Grok 5 will reportedly be natively multimodal across text, images, audio, and real-time video, but capability on paper doesn't win deals that have already been signed.”
To keep momentum, xAI has been shipping around the gap: Grok Voice rolled out publicly in early June, and Grok Imagine Video reached preview with API access shortly after. It's a sensible holding pattern, but the company's position in the model race ultimately rests on whether the 6-trillion-parameter bet trains -- and serves -- economically.