Image: xAI / x.ai Grok 4.6 Is Coming This Week: What's Changing and What Isn't
xAI is releasing Grok 4.6 around August 7, with Grok 4.7 (the actual 2.1T model) following weeks later. Both matter for Cursor users now that SpaceX owns Anysphere.
xAI has two model releases stacked up for August, and Elon Musk confirmed the timeline on SpaceX’s August 4 earnings call. Grok 4.6 lands around August 7. Grok 4.7, the bigger one, follows a few weeks after that.
If you’re using Cursor, this matters. SpaceX completed its acquisition of Anysphere earlier this year, and Grok 4.5 was already integrated into Cursor as a first-party model option. The next two Grok releases will presumably follow the same path.
What Grok 4.6 actually is
This is where a lot of the coverage has been wrong. Grok 4.6 is not a larger model. It uses the same 1.5 trillion parameter foundation as Grok 4.5. Musk specifically described it as “the 1.5T model with significantly improved SFT and RL,” meaning the gains come entirely from post-training: better supervised fine-tuning and reinforcement learning.
What that typically translates to in practice: more consistent instruction-following, fewer hallucinated tool calls, better performance on agentic tasks that require multi-step reasoning, and reduced tendency to pad outputs. All of that matters for coding workflows.
Because the parameter count is unchanged, serving costs shouldn’t change either. Grok 4.5 runs at $2 per million input tokens and $6 per million output tokens in the xAI API. Grok 4.6 is likely to land at the same price.
Grok 4.7 is the bigger release
Grok 4.7 is where the actual scale jump happens. Musk described it as “the 2.1T model, better than 4.6 in every way, except slightly slower to serve.” That 40% parameter increase over the 1.5T model should produce meaningful capability jumps, though slower inference means it will probably carry a price premium.
No confirmed date yet, just “a few weeks” after 4.6.
Why this release cadence matters
xAI has been moving fast. Grok 4.5 came out in July, 4.6 comes this week, 4.7 a few weeks after that. The underlying reason is infrastructure: xAI’s Colossus cluster in Memphis has crossed 500,000 GPUs across multiple facilities, with a target of one million. That kind of compute makes rapid iteration practical.
The pace also reflects competitive pressure. Claude Opus 5, which Anthropic released in July, runs at roughly half the cost of its predecessor with improved coding and agentic performance. OpenAI has been cutting prices aggressively on its GPT-5 variants. xAI is keeping up by shipping faster rather than taking big swings.
What Cursor users should expect
Grok 4.5 integration in Cursor happened quickly after the SpaceX acquisition. There’s no reason to expect 4.6 and 4.7 to be different, though the actual timing of when they show up in the model picker hasn’t been confirmed.
If you’re building with Cursor’s API access rather than the editor directly, the practical advice is what it always is before a model drop: keep your model selection configurable and maintain fallbacks so you can test on 4.6 without rewriting anything.
The SFT and RL improvements in 4.6 should make it noticeably better than 4.5 for the kinds of tasks where 4.5 sometimes fails to follow complex instructions or drops context mid-agent-run. Those are the edges that usually improve most from post-training work.
Benchmarks will tell us more once the model is live.