Elon Musk Says Grok 4.5 AI Model Is Now in Private Beta at SpaceX and Tesla

If you use a Tesla, follow AI news, or work in software development anywhere in Kent, you’ll have noticed that the race between the big AI labs is moving faster than most people can keep up with. The latest move comes from Elon Musk, who has announced that xAI’s newest large language model — Grok 4.5 — is now being tested privately inside SpaceX and Tesla. It isn’t available to the public yet, but the claims being made about it are striking enough to be worth paying attention to.

Musk posted the announcement on his verified @elonmusk account on X, stating that Grok 4.5 is currently in private beta at his two companies. That means engineers and staff at SpaceX and Tesla are using it internally, but no one outside those organisations can access it yet.

What Is Grok 4.5, Exactly?

Grok is the family of AI language models developed by xAI, the artificial intelligence company Musk founded as a rival to OpenAI, Anthropic, Google DeepMind and others. Previous versions of Grok have been integrated into the X platform — the social media site formerly known as Twitter — but Grok 4.5 appears to be a more powerful internal frontier version aimed at advanced enterprise use within Musk’s own companies.

According to Musk’s post, Grok 4.5 is built on what xAI calls the V9 foundation model, with a claimed parameter count of around 1.5 trillion. Parameters are, roughly speaking, the numerical settings inside an AI model that determine how it processes and generates information — the more you have, the more complex the patterns the model can theoretically learn. For context, Grok 4.4, the previous version, was widely reported to have around one trillion parameters, suggesting this is a meaningfully larger model.

Musk also says the model received supplemental training using data from Cursor, an AI-powered coding environment popular with software developers. That detail matters because it suggests Grok 4.5 has been specifically tuned for software engineering tasks — writing code, refactoring it, and handling what the industry calls “agentic” workflows, where an AI model takes sequences of actions rather than just answering a single question.

The Performance Claims — and the Caveats

Musk’s post says early internal evaluations show Grok 4.5 performing “close to, perhaps exceeding Opus” — a reference to Claude Opus, Anthropic’s large language model and one of the most capable publicly available AI systems. He also states that reinforcement learning is continuing to improve the model, and that the associated Grok Build framework — xAI’s coding and agent toolkit — is “getting better every day.”

Those are bold claims. But they come entirely from internal evaluations at xAI, SpaceX and Tesla. No independent benchmarks, peer-reviewed assessments or external access to Grok 4.5 have been published. Without that, it’s genuinely difficult for anyone outside those organisations to verify what the model can and can’t do, or how it compares to Claude Opus in practice.

That gap matters. Several technology analysts have pointed out that self-reported performance figures from AI labs, without independent testing, tell only part of the story. There’s also no public information yet about Grok 4.5’s safety guardrails, training data composition, or how it handles potential harms — all things that regulators and civil society groups have increasingly been pushing AI companies to document openly.

A Monthly Model Release Cadence?

Perhaps the most eyebrow-raising part of Musk’s announcement is the claim that SpaceX plans to release completely new, trained-from-scratch AI models every single month this year. That would represent an extraordinarily fast pace of development — most frontier AI models take many months or longer to train and evaluate.

Whether that cadence is realistic, and whether it can be matched with thorough safety testing at each stage, is unclear. No external logs or publications currently corroborate the plan. But if it holds, it would suggest xAI is pursuing a strategy of rapid, iterative experimentation — testing different architectures, data mixes and training approaches at a pace that few other labs have publicly committed to.

The Wider AI Race

xAI’s push with Grok 4.5 sits inside a broader and intensifying competition among AI labs. OpenAI, Anthropic, Google DeepMind and Meta are all developing and releasing frontier models at pace, each claiming advances in reasoning, coding, and the ability to handle complex, multi-step tasks. Governments and regulators — including in the UK — are watching this acceleration closely, with growing pressure on companies to submit their most powerful models to independent safety evaluations before or alongside public release.

The UK’s AI Safety Institute, based in London, has been working with major labs on exactly this kind of testing. Whether Grok 4.5 or future xAI models will be submitted to such processes isn’t yet known.

Gary Marcus, an AI researcher and critic who has consistently called for greater transparency from AI developers, said: “We need independent evaluations of these systems, not just company announcements. The history of AI is littered with overclaimed benchmarks.”

What This Means for Kent Residents

For people here in Kent, Grok 4.5 isn’t something you can use or interact with right now — it’s locked inside SpaceX and Tesla’s internal systems. But if and when xAI opens access more broadly, it could eventually find its way into Tesla vehicles on Kent roads, developer tools used by local software firms, or the X platform that many residents use daily. Any UK deployment would need to comply with UK GDPR and consumer protection law, and Kent-based developers may find it worth watching whether xAI follows through on promised public access to Grok’s coding capabilities later this year.