Elon Musk doesn’t do things in moderation, and his AI company xAI is no exception. Barely two weeks after launching Grok 4.5 — itself a significant model trained alongside Cursor for coding and agentic tasks — Musk has confirmed that Grok 4.6 will arrive around August 7, with an even larger Grok 4.7 following “a few weeks” after that.
The announcement, made via a post on X on July 28, puts xAI on track to release three frontier-class models in under two months. Whether that pace is sustainable or just Musk being Musk remains to be seen.
What We Know About Grok 4.6
Grok 4.6 is described as a 1.5-trillion-parameter model — a substantial step up from Grok 4.5. The key upgrades are in supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF), the two post-training techniques that turn a raw language model into something actually useful.
In plain terms: the base model is bigger, and the polish on top of it should be considerably better. Musk hasn’t shared benchmarks yet, but given that Grok 4.5 already scores competitively against GPT-5 and Claude Opus 4 on coding and reasoning tasks, 4.6 should push those numbers further.
The target date is “around August 7,” which in Musk time could mean August 7 or could mean August 14. Plan accordingly.
Grok 4.7: The 2.1-Trillion-Parameter Follow-Up
If 1.5 trillion parameters felt excessive, xAI would like you to hold its beer. Grok 4.7 is pegged at 2.1 trillion parameters, and Musk has described it as “better than our 1.5T in every way.” No specific launch date has been given beyond “a few weeks” after 4.6, which puts it somewhere around late August or early September.
That’s a staggering number. For context, the original GPT-3 that kicked off the modern AI era had 175 billion parameters. We’re now casually discussing models more than 10 times that size.
The Hardware Making This Possible
None of this happens without compute, and xAI has been spending aggressively. The company’s Colossus cluster in Memphis, Tennessee — already one of the largest AI training facilities on the planet — currently houses over 500,000 GPUs across its Colossus 1 and Colossus 2 buildings. A third facility in nearby Southaven, Mississippi is being converted, and xAI has publicly stated its ambition to reach one million GPUs.
The AI compute arms race is getting expensive. Nvidia recently confirmed that Colossus is also receiving an additional 100,000 Hopper GPUs. When your GPU orders need their own press releases, you’re operating at a different scale.
Why the Breakneck Pace?
The AI landscape right now is essentially a three-way sprint between OpenAI (GPT-5.5), Anthropic (Claude Opus 4.8), and xAI (Grok 4.x). Google’s Gemini is in the mix too, but Musk’s focus seems squarely on outpacing his former OpenAI colleagues.
xAI’s strategy is clear: ship fast, iterate in public, and let raw scale do the talking. Grok 4.5 was the company’s first model built specifically for agentic tasks and coding. If 4.6 and 4.7 deliver on their parameter counts and training improvements, xAI will have gone from competitive to potentially leading in frontier AI performance — all in a single summer.
How to Access Grok Today
Grok is available through several tiers, from limited free access to the $300/month SuperGrok Heavy plan. Here’s the current breakdown:
| Plan | Monthly Price | What You Get |
|---|---|---|
| Free (X / grok.com) | $0 | Limited Grok access |
| X Premium | $8 | Higher Grok usage limits |
| SuperGrok Lite | $10 | Standalone Grok, moderate limits |
| SuperGrok | $30 | Full Grok access, higher limits |
| X Premium+ | $40 | X features + expanded Grok |
| SuperGrok Heavy | $300 | Maximum limits, priority access |
How quickly Grok 4.6 and 4.7 roll out across these tiers is anyone’s guess. Grok 4.5 took about two weeks to go from API launch to the mobile apps, so expect a similar staggered rollout.
Should You Care?
If you’re already using AI tools for work or development, yes. The gap between frontier models is shrinking, and xAI shipping two major upgrades before September means the competitive pressure on OpenAI and Anthropic to respond is real. That’s good for everyone — faster iteration means better tools, sooner.
If you’re just casually using Grok through X, you probably won’t notice the version number change. You’ll just notice it getting better at answering your questions. Which is sort of the point.
How many parameters does Grok 4.6 have?
Grok 4.6 is a 1.5-trillion-parameter model, up from the previous Grok 4.5. It features upgraded supervised fine-tuning and reinforcement learning from human feedback.
When does Grok 4.6 release?
Elon Musk said Grok 4.6 will release “around August 7, 2026.” xAI timelines are approximate, so a few days’ variance is possible.
What about Grok 4.7?
Grok 4.7 is a 2.1-trillion-parameter model expected “a few weeks” after Grok 4.6, likely late August or early September 2026. No exact date has been confirmed.
How does Grok compare to ChatGPT and Claude?
Grok 4.5 already benchmarks competitively with GPT-5 and Claude Opus 4 across coding, math, and reasoning tasks. Grok 4.6 and 4.7 are expected to push those scores higher, though benchmarks haven’t been released yet.
Leave a Reply