Is Grok 5 Slipping Again? Inside xAI's Expanding Roadmap — and Abu Dhabi's Stake in the Outcome

Elon Musk has been promising Grok 5 for almost a year now, and the finish line keeps moving. The latest wrinkle, reportedly laid out by Musk himself in mid-September, is that two more models — Grok 4.8 and Grok 4.9 — have reportedly been slotted in ahead of it. Nothing here is official from xAI in the form of a model card or release date, so treat the whole picture as a moving target rather than a confirmed roadmap. But it's worth tracking closely in the Gulf specifically, because Abu Dhabi's MGX fund is reportedly among the investors bankrolling whatever xAI ships next.

A deadline with a long history of slipping

Musk first floated a Grok 5 release "before the end of 2025" in comments made in mid-2025. That target reportedly slid to early 2026, then to a first-quarter window, then to the second quarter — none of which materialized. According to a CometAPI report citing Musk's remarks on SpaceX's August 2026 earnings call, he pushed the timeline again, this time saying Grok 5 was planned for "before the end of 2026" and claiming — without elaboration — that the model would be trained in part on roughly 25 years of SpaceX flight and engineering data. None of that has been independently verified, and xAI has not published a model card, benchmark results, pricing, context-window specs, or parameter counts for Grok 5. At this point, specifics are coming almost entirely from Musk's own social posts and earnings-call asides rather than from xAI's own documentation.

Two models cut in line

The latest twist, as reported by outlets tracking Musk's September 14 posts, is that Grok 5 now appears to be queued behind not one but two intermediate releases. Grok 4.8 is reportedly a roughly 2.5-trillion-parameter model built on a new internal C++ training stack, which Musk said was finishing its base training and moving into reinforcement learning "this week." Grok 4.9 is reportedly being positioned as something closer to a mid-tier competitive release — in the same rough performance class as rivals' current flagship models — with no timing attached. Musk is said to be reserving the "Grok 5" label specifically for a model he's framed as an AGI-level leap, which is itself a claim that should be read with heavy skepticism given how loosely that term gets used across the industry.

It's worth noting that none of these claims about 4.8 actually completing training, entering RL, or shipping on schedule has been confirmed by xAI through any official channel as of this writing. Given that Grok 4.7 was apparently still the newest shipped model as recently as late September, according to tracking sites, the 4.8-then-4.9-then-5 sequence described by Musk amounts to at least two more unconfirmed launches standing between today and whatever Grok 5 turns out to be.

Why Abu Dhabi has skin in this game

This is where the story gets a Gulf angle. In January 2026, xAI closed a Series E funding round reportedly worth around $20 billion, exceeding its original $15 billion target. According to reporting from AGBI (Al Gharbia Business Insight), Abu Dhabi's MGX fund and a Qatari sovereign wealth vehicle were both participants, alongside Nvidia and Cisco Investments, though the exact size of each investor's contribution was not disclosed. MGX — founded in 2024 by Abu Dhabi's G42 and Mubadala — has reportedly also taken positions in OpenAI and Anthropic, making it one of the more diversified sovereign bets across the major AI labs rather than a wager tied to any single one.

xAI has said the Series E proceeds are earmarked for infrastructure build-out, compute, and product deployment — which matters here because it ties Abu Dhabi's capital directly to the compute and data-center capacity that any Grok 5-class model would actually need to train and serve. If the roadmap keeps slipping, it's not just an inconvenience for Musk's timeline; it's a live question for how quickly that infrastructure spend converts into a shipped product that Gulf-based sovereign investors are counting on.

The skepticism is earned

It's worth contrasting this with how OpenAI handled its own last major release: GPT-5.6 was previewed to a limited set of organizations in June 2026 and then rolled out broadly in July, roughly on the schedule OpenAI had signaled, with an actual model card, documented pricing tiers, and published benchmark comparisons. Grok's pattern looks different — a string of dates announced on social media and earnings calls, several of which have already come and gone without a model shipping, and a growing queue of interim releases between "now" and "Grok 5."

None of this means Grok 5 won't ship, or that it won't be capable when it does. But the available evidence right now is: no confirmed date, no technical specification, two unreleased models reportedly ahead of it in the queue, and a public track record of missed self-imposed deadlines. Readers should treat every number attached to Grok 5 — parameter count, training data, release window — as unverified until xAI publishes something with its own name on it.

What to watch next

The nearer-term signal worth following is whether Grok 4.8 actually finishes reinforcement learning and ships in the coming weeks, which would at least suggest the roadmap Musk described in September is holding up. Until then, Grok 5 remains a rumor with a widening gap between what's been promised and what's been delivered — one that a slice of Abu Dhabi's AI investment strategy is now indirectly riding on.