The toughest verdict on Grok 4.7 came from the person least likely to dodge the question. On September 27, xAI founder Elon Musk admitted in an interview with China Media Group (CMG) that Grok 4.7 is not as good as Anthropic's newly released Opus 5.5 — his exact words were "a solid workhorse of a model," but "not as good as, say, Opus 5.5 that just got released." In the interview, conducted at Tesla's global engineering headquarters, Musk unusually quantified xAI's gap with its rivals: xAI has been seriously working on AI for about three years, Anthropic for about six, and he expects xAI to "catch up to frontier sometime next year most likely."
What he actually said
Beyond conceding the gap, Musk disclosed several business signals that had not been made public before:
- Grok Bot is growing fast: xAI's personal digital assistant, Grok Bot, is "growing about 100% a month" — one of the fastest-growing product lines he chose to highlight.
- A data strategy pivot toward the real world: xAI has so far used "only a little bit" of SpaceX's training data and plans to bring in Tesla's training data in the future, with the goal of making AI good at "real world engineering." By contrast, Anthropic's strength, in his framing, is making AI good at software engineering.
- Praise for Chinese models: he described Chinese AI models as excellent, with the best performance per unit of compute in the world, and predicted China will resolve its lithography and chip-compute constraints within two to three years.
The interview also covered Cybercab's commercial operations in Texas, a prediction of more than one billion humanoid robots within ten years, and "universal high income" — classic long-range Musk visions, carrying less concrete information than the three points above.
How this differs from his usual tone
Musk has typically spoken about Grok in an all-out offensive posture. Openly admitting it trails Opus 5.5, and pinning the catch-up to "next year," is essentially expectation management: put the short-term shortfall on the table first, then redirect the story to xAI's proprietary data moat.
The timeline is worth noting. Opus 5.5 was released on September 22 and immediately topped the Text Arena leaderboard (see this column's September 27 AI briefing); Grok 4.7 is the flagship of the accelerated release cadence that began in late July, and an earlier evaluation found it could break into the top four in agentic coding — at a cost of roughly 80,000 tokens per task. Musk's "solid workhorse" label matches that profile closely: capable and dependable, but not the strongest.
Three signals behind the concession
First, the competitive focus is shifting from "whose model is smarter" to "whose data no one else can get." Anthropic is turning AI into a software-engineering specialist; xAI wants to turn the real physical data accumulated from rockets, factories, and autonomous driving into a moat. The catch, as Musk himself admitted, is that "only a little bit" of SpaceX's data has been used so far — the narrative is still at the blueprint stage, and the bet placed when Grok 4.5 entered private testing at SpaceX and Tesla still needs follow-through investment to pay off.
Second, Grok Bot's 100%-a-month growth shows xAI tilting toward consumer personal assistants. Frontier labs are all hunting for growth curves beyond the model itself: Meta has recently been pairing hardware with its personal agent too. For ordinary users, Grok's future experience gains may come more from product forms like the Bot than from the base model.
Third, the takeaway for developers choosing models is direct. For tasks that need frontier capability, Opus 5.5 remains the safer choice; Grok 4.7 is positioned as the "dependable workhorse," suited to cost- and stability-sensitive scenarios where extreme capability is not the top requirement. Only when xAI actually puts real-world engineering data to work could this judgment change.
