Musk Says Grok 4.6 Expected to Launch Next Week
Elon Musk stated that xAI's Grok 4.6 is likely to be released next week, having previously indicated a timeframe in early August.
Public information shows that Grok 4.6 is expected to have approximately 1.5 trillion parameters, with significant improvements in supervised fine-tuning (SFT) and reinforcement learning (RL); its subsequent version, Grok 4.7, is described as larger but slower in response and more efficient in token usage.
This pace continues xAI's high-frequency iteration strategy, with the release interval between Grok 4.6 and 4.7 compressed to a few weeks, demonstrating its ongoing pursuit of "faster releases, stronger parameters, and higher efficiency" to compete with leading models.
Source: Public Information
ABAB AI Insight
The update logic of Grok 4.6 indicates that xAI is not only competing on benchmark scores but also on the rhythm of "training completion—release—retraining"; in the competition of leading models, the release time itself is part of product strength.
Similar cases can be compared to OpenAI and Anthropic's strategy of continuously shortening model iteration cycles over the past year: the shorter the intervals between models, the harder it is for the market to form stable expectations, increasing user migration costs and accelerating the depreciation of older models' value.
On the capital path, xAI is forming a "fast first, then large" dual-model route with Grok 4.6 and 4.7: 4.6 emphasizes usability and efficiency, while 4.7 emphasizes higher parameters and lower token costs. This layered approach helps to divert developers and subscription users into different price tiers.
Essentially, this represents a technological substitution: AI model competition is shifting from "one-time release wins" to "continuous iteration capability," with the underlying mechanism being that marginal improvements in model performance increasingly rely on post-training, computational power, and data integration, rather than one-time parameter expansion.
ABAB News · Cognitive Laws
The moat of models is becoming the release rhythm.
Large parameters do not necessarily mean faster, but faster iterations are certainly more expensive.
The ultimate goal of AI competition is not a single victory, but to remain continuously competitive.