The list of flagship AI models that missed their promised launch windows continues to grow.

Last summer, billionaire Elon Musk, the founder and CEO of AI company xAI, said that Grok 3, xAI’s next major AI model, would arrive by “end of year” 2024. Grok, xAI’s answer to models like OpenAI’s GPT-4o and Google’s Gemini, can analyze images and respond to questions, and powers a number of features on X, Musk’s social network.

“Grok 3 end of year after training on 100k H100s should be really something special,” Musk wrote in a July post on X, referring to xAI’s enormous Memphis-based cluster of GPUs. “Grok 3 will be a major leap forward,” he said in a mid-December follow-up post.

Yet it’s January 2, and Grok 3 hasn’t arrived — nor are there signs its rollout is imminent.

In fact, some code on xAI’s website spotted by AI tipster Tibor Blaho suggests that an intermediate model, “Grok 2.5,” may land first.

Grok[.]com is possibly coming soon with Grok 2.5 model (grok-2-latest – “Our most intelligent model”) – thanks for the hint, anon! pic.twitter.com/emsvmZyaf7

— Tibor Blaho (@btibor91) December 20, 2024

Granted, this isn’t the first time Musk has set a lofty goal and missed it. It’s well-established that Musk’s pronouncements about the timing of product launches are often unrealistic at best.

But Grok 3’s MIA status is interesting because it’s part of a growing trend.

Last year, AI startup Anthropic failed to deliver a successor to its top-of-the-line Claude 3 Opus model. Months after announcing that a next-gen model, Claude 3.5 Opus, would by released by the end of 2024, Anthropic scrapped all mention of the model from its developer documentation. (According to one report, Anthropic did finish training Claude 3.5 Opus sometime last year, but decided that releasing it didn’t make economic sense.)

Reportedly, Google and OpenAI have also suffered setbacks with their flagship models in recent months.

To blame for all this, in large part, are the limits of current AI scaling laws — the methods companies are using to increase the capabilities of their models. In the not-too-distant past, it was possible to achieve substantial performance boosts by training models using massive amounts of computing power and larger and larger data sets. But the gains with each generation of model have begun to shrink, leading companies to pursue alternative techniques.

There could be other reasons for Grok 3’s delay. xAI has a much smaller team than many of its rivals, for one. Nonetheless, the slipped launch time frame adds to the body of evidence that conventional AI training approaches are running up against a wall.