Grok 4.3 beta rolls out
xAI pushed Grok 4.3 into beta — a 1‑trillion‑parameter model described as roughly twice the size of Grok 4.20 with longer training — first to SuperGrok Heavy tier users. (x.com) xAI said broader rollout to all SuperGrok subscribers is coming soon and the announcement highlighted expected performance gains. (x.com)
xAI has started rolling out Grok 4.3 beta to SuperGrok Heavy subscribers, adding a new test version of its flagship chatbot to its highest-paid tier first. (piunikaweb.com) A parameter is a tuning value inside an artificial intelligence model, and xAI described Grok 4.3 beta as a 1 trillion-parameter system, about twice the size of Grok 4.20, with a longer training run. xAI said broader access for all SuperGrok subscribers is coming soon. (piunikaweb.com) Grok is xAI’s chatbot product, built into X and also sold through subscriptions and an application programming interface. xAI launched Grok 4 and the $300-a-month SuperGrok Heavy plan on July 9, 2025. (x.ai; teslarati.com) The company has framed Grok’s recent upgrades around scale and tool use rather than a brand-new product line. In its July 2025 launch post, xAI said Grok 4 used native tools such as web search and code execution, and said training ran on its Colossus cluster of 200,000 graphics processing units, or GPUs. (x.ai) That makes this beta release less about a new app than about xAI’s release strategy: big model updates are landing first behind a premium paywall, then moving down to lower-paid subscribers later. The same pattern appeared when xAI introduced SuperGrok Heavy as the first home for Grok 4 Heavy. (x.ai; piunikaweb.com) The size claim also fits a broader race among artificial intelligence companies to signal progress with bigger models, longer training, and stronger benchmark scores. In its Grok 4 announcement, xAI highlighted performance on Humanity’s Last Exam, a difficult test of expert-level questions, and compared Grok 4 against OpenAI and Google models. (x.ai) xAI has not published a full standalone technical paper for Grok 4.3 beta in the sources reviewed here, so outside verification of the new model’s architecture and performance is still limited. The company’s public materials for Grok 4 emphasized benchmark charts, tool use, and training scale rather than detailed model weights or methods. (x.ai; piunikaweb.com) For users, the immediate change is simple: Grok 4.3 beta is live now on xAI’s top subscription tier on web and mobile, while standard SuperGrok customers are waiting for the wider rollout xAI said is next. (piunikaweb.com)