SAN FRANCISCO — xAI has started rolling out Grok 4.3 to SuperGrok and Premium+ subscribers, the company confirmed, delivering an updated pre-trained model that brings architectural improvements and a refreshed knowledge cutoff without increasing the parameter count beyond Grok 4.20's scale.
The release continues xAI's pattern of iterative model updates that ship improvements ahead of major version releases — keeping its subscriber base on the frontier while the larger Grok 5 project continues in training.
What Is New in Grok 4.3
Grok 4.3 is described by xAI as a new pre-trained model at the same scale as Grok 4.20, but with an improved architecture and a December 2025 knowledge cutoff — a meaningful refresh given how quickly the AI landscape has evolved over the past six months.
The improved architecture is aimed at delivering better instruction-following and more consistent reasoning performance across long-context tasks, an area where earlier Grok 4 variants showed occasional inconsistencies. The updated knowledge cutoff brings the model current on developments in AI, technology, geopolitics, and science through the end of last year.
SuperGrok and Premium+ subscribers can select Grok 4.3 explicitly in the model picker on grok.com, X, and the iOS and Android apps, or access it automatically in Auto mode as xAI's recommended option for demanding tasks.
Where Grok 4.3 Fits in the Lineup
Grok 4.3 sits alongside Grok 4.1 in xAI's current model catalog. Grok 4.1 remains available to all users, while 4.3 is positioned as a premium option for subscribers who want the most capable version currently available.
The update comes as xAI continues preparations for Grok 5, which the company has described as a step-change in capability built on dramatically more parameters and a new training approach. Grok V9 Medium — at 1.5 trillion parameters, three times the scale of current Grok 8 small — completed training in late May, with a public release targeting mid-June 2026.
The Broader xAI Momentum
The Grok 4.3 release arrives as xAI is deploying on multiple fronts. Grok Build, the company's agentic coding assistant, shipped version 0.2.11 in late May and has been adopted by Wall Street firms including Apollo and Morgan Stanley for internal testing. Grok Voice is now handling Starlink customer support calls. Colossus 2, xAI's next-generation training cluster, is running seven models in parallel.
The pace of iteration at xAI has accelerated meaningfully in 2026. With Grok 4.3 now in subscribers' hands and Grok 5 approaching, xAI is positioning itself to hold ground against OpenAI, Google, and Anthropic — all of which have accelerated their own release cadences in the same period. For users on the platform today, 4.3 represents the strongest version of Grok available, with better architecture than what came before and a knowledge base that reflects the world as it stood six months ago.
