Grok 4.6 Update: Elon Musk has revised the numbers on his own upcoming AI models. Here’s what’s different from what was expected just days ago.
xAI founder Elon Musk has posted fresh details on X about Grok 4.6 and Grok 4.7, giving both a firmer timeline and updated specs. According to Musk, Grok 4.6 will release around August 7, built as a 1.5-trillion-parameter model with “significantly improved SFT & RL” — that’s supervised fine-tuning and reinforcement learning, the post-training techniques that shape how a raw model actually behaves once it’s talking to users. Grok 4.7 will follow a few weeks later as a larger 2.1-trillion-parameter model.
This is a direct update to Grok’s release story TechnoSports covered earlier this week — worth noting because the numbers have shifted since then, a reminder of how fluid these pre-launch announcements can be.
Table of Contents
What Changed From Earlier Reports
| Detail | Earlier Reports | Musk’s Latest Post (July 28) |
|---|---|---|
| Grok 4.6 parameters | ~2 trillion | 1.5 trillion |
| Grok 4.6 release window | “Weeks away” | Around August 7 |
| Grok 4.7 parameters | Larger than 4.6 (unspecified) | 2.1 trillion |
| Grok 4.7 timing | ~4 weeks after 4.6 | “A few weeks” after 4.6 |
The takeaway: Grok 4.6 is landing smaller than initially estimated, while Grok 4.7 is confirmed as the genuinely large jump, at 2.1 trillion parameters.
What Musk Says Each Model Improves
Grok 4.6’s headline change is about polish rather than raw scale — better SFT and RL should mean sharper instruction-following and more reliable reasoning, even without a huge size increase. Grok 4.7, on the other hand, is being positioned as a strict upgrade over 4.6 “in every way,” with one tradeoff: Musk says it will be slightly slower to serve, offset by better token efficiency, meaning it should still be more cost-effective to run per response despite the added size.
The Context Behind the Announcement
Musk’s post came in response to a benchmark from Vercel CEO Guillermo Rauch, who found Grok 4.5 to be the best-performing cybersecurity AI model on a price-performance basis in Vercel’s latest testing — cheaper than OpenAI’s GPT-5.6 Sol, Anthropic’s Claude Opus 5, and Moonshot’s Kimi K3, while matching Kimi-level performance. Rauch still ranked Sol ahead as the overall frontier leader. That comparison appears to have prompted Musk’s “interesting” reply teasing what’s coming next.
Still No Independent Verification
As with the earlier report, everything here comes directly from Musk’s own posts rather than official xAI documentation, benchmarks, or technical papers. No pricing, context window, or API details have been shared yet. One outlet cautioned that “around August 7” in Musk’s typical announcement style could realistically stretch to mid-August — worth keeping in mind before circling a calendar date, a pattern TechnoSports has seen play out with past xAI announcements too.
Bottom Line
The core story hasn’t changed — xAI is still moving at an unusually fast release cadence, aiming to ship two frontier-class models within weeks of each other. What has changed is the specific numbers: Grok 4.6 is smaller than first estimated, and Grok 4.7 now has a confirmed 2.1-trillion-parameter figure attached to it. Treat both as Musk’s own stated targets rather than locked-in facts until xAI publishes something official.
Based on Elon Musk’s official posts on X, corroborated by Reuters-linked coverage and multiple tech outlets. xAI has not released independent technical documentation or benchmarks for either model.
