Grok 4.5 enters private beta with 1.5T parameters

Elon Musk just revealed that xAI's Grok 4.5 has entered private beta, landing directly on the workstations of engineers at SpaceX and Tesla.

A 1.5T parameter playground

According to Musk, Grok 4.5 is built on their 1.5 trillion parameter V9 foundation model, supplemented by training on data from AI code editor Cursor.

Musk claimed that early internal evaluations show Grok 4.5 performing close to, or perhaps even exceeding, Anthropic's Claude Opus. Reinforcement learning is still improving the model significantly, alongside daily upgrades to the Grok Build harness.

He also made another aggressive promise: SpaceX will release completely new models trained from scratch every single month this year.

Why it matters

External developers cannot touch Grok 4.5 just yet, but using the engineering talent across Tesla and SpaceX as an active testing ground is a massive feedback loop.

We think folding Cursor data into the mix clearly shows xAI wants to dominate developer workflows and software engineering, taking direct aim at Claude Opus's reputation as the top coding assistant.

Still, rolling out ground-up new models every month sounds wildly demanding from both a compute and validation standpoint. Whether this rapid release cadence produces meaningful leaps or just marketing noise will depend entirely on how the public versions hold up under pressure.