xAI’s latest model isn’t going to the public. It’s going to work.
Grok 4.5, announced on June 28 via Elon Musk’s X account, has entered a private beta testing phase exclusively with teams at SpaceX and Tesla. The model is built on xAI’s new V9 foundation, which packs 1.5 trillion parameters, making it roughly three times larger than the V8-small architecture that powered earlier Grok 4 variants.
What’s under the hood
xAI supplemented its training pipeline with Cursor data, a move specifically designed to sharpen coding and technical competencies. Cursor is a popular AI-powered code editor.
Early internal evaluations suggest Grok 4.5’s performance is comparable to, if not superior to, Anthropic’s Claude Opus.
Musk has also referenced advancements tied to reinforcement learning features and the Grok Build system, an internal tooling framework. Real-world applications inside SpaceX and Tesla are reportedly feeding back beneficial improvements, creating a loop where the model gets better by actually doing useful things rather than just training on static datasets.
The private beta strategy
xAI has not specified a public release date for Grok 4.5 or outlined broader accessibility plans. The original timeline had pegged the model for a late May 2026 launch, meaning it’s already running behind schedule by about a month.
Grok 4 itself only launched in July 2025. Going from Grok 4 to a 1.5 trillion parameter successor in under a year is an aggressive pace.
xAI’s roadmap includes monthly model releases throughout 2026, with variants of Grok 5 projected to reach 10 trillion parameters.