Grok 4.5 is now running inside SpaceX and Tesla in a closed test. It is not available to the public yet, and no release date has been confirmed. But the announcement tells you quite a bit about where Musk's AI operation is heading.
The model is built on a new foundation that is roughly three times larger than the one currently powering Grok on the X platform. To put that in plain terms: the version of Grok that most users interact with today is running on the older, smaller model. Grok 4.5 is the next step up, and it is still in internal testing.
The Cursor connection matters more than it might appear. SpaceX agreed to buy Cursor, a widely used AI coding tool with around $2.6 billion in annual revenue, for $60 billion in stock just two weeks ago. That deal has not officially closed yet, but data from Cursor's coding tool was already used to train Grok 4.5. The model is now being tested on real engineering work at SpaceX and Tesla before anyone else sees it. Most AI companies test their models on artificial benchmarks. xAI is testing on actual rocket engineering and automotive software development pipelines.
Musk's claim that Grok 4.5 matches or beats Claude Opus deserves some caution. Claude Opus 4.8, currently the top-ranked model on independent leaderboards, leads on most published coding and reasoning tests by a documented margin. Grok's previous versions were competitive on cost and speed but trailed Anthropic on the hardest, most complex tasks. Grok 4.5 has not yet been tested publicly or independently. What Musk is citing are internal evaluations, not third-party scores.
The plan to release completely new models every month for the rest of 2026 is genuinely unusual. Most AI labs take months between major releases. xAI is signalling that it will use its SpaceX computing infrastructure to keep pushing out new versions at a pace competitors will struggle to match.
There is, however, a real obstacle that faster model releases do not automatically solve. Grok has had a difficult year on safety and reliability. Multiple regulatory investigations are currently active: an EU probe launched in January, a UK regulator inquiry, and ongoing concerns from US federal agencies who flagged the model as unreliable for government use. A former xAI engineer filed a lawsuit in June alleging he was fired for raising safety concerns, and a February wave of departures from the company included senior safety staff. An independent analyst told CBS News that Grok has so far "failed to make a dent in the frontier market," which remains led by Anthropic, OpenAI, Google, and Meta in that order.
For any organisation currently evaluating AI tools for serious internal use, the picture with xAI is this: the raw model capability is improving fast, and the cost-per-use is significantly lower than Anthropic or OpenAI. But the safety track record, the regulatory exposure, and the relatively short history of its enterprise offering are all real factors to weigh before committing.
The monthly release cadence is worth watching. If xAI can sustain it and the quality holds up through independent testing, the competitive pressure on Anthropic and OpenAI will increase through the second half of 2026. If it cannot, the announcements become noise. The market will have a clearer answer by the end of the year.