Describing Grok 4.7 as a 'solid workhorse of a model,' Musk acknowledged that Anthropic, which has spent longer developing frontier models, holds the technological lead for now.
'I would expect that we probably catch up to frontier, sometimes, next year, most likely,' he said, adding that Grok remains 'very useful for many things' in the meantime.
A gap measured in years
Musk attributed the performance gap partly to how long each company has been in the field. SpaceXAI's AI efforts are roughly three years old, he said, while Anthropic has been building models for about six years — an experience gap that shows in benchmark-leading systems like Opus 5.5, which Anthropic released on September 22.
The comments echo earlier remarks from mid-September, when Musk said even the unreleased Grok 4.8 — built on a new software stack with 2.5 trillion parameters — would improve significantly but still not match OpenAI's Astra or Anthropic's Fable-class models.
Real-world engineering, not software
Rather than competing head-on in software engineering, Musk said SpaceXAI is pursuing a different frontier: making AI 'excellent at real world engineering.'
Anthropic's products, he said, are focused on making AI excellent at software engineering. Grok, by contrast, could draw on unique physical-world data from Tesla's vehicles and SpaceX's rockets and spacecraft to understand the physical world in ways rivals cannot.
He also highlighted Grok Bot, SpaceXAI's agentic AI personal assistant, saying it has been 'very popular and is growing about a 100% a month' — one of the fastest adoption curves the company has seen.
Praise for China's AI push
In the wide-ranging interview, Musk praised Chinese manufacturing, the compute efficiency of Chinese AI models, and Tesla's Shanghai Gigafactory. He predicted China could overcome lithography and chip manufacturing limitations within about three years.
The remarks landed just as China's AI labs race to close the compute gap with the West — a shift that, combined with Musk's own catch-up pledge, suggests the AI leaderboard could look very different by 2027.
