Grok 4.7 vs Grok 4.6
Grok 4.7 replaced Grok 4.6 as xAI's frontier model, and the two are close enough in behaviour that the differences are easy to miss in a short chat. They are easier to see once you push on reasoning, fresh information, or image work.
Here is the short version of what changed, and when the older model is still the sensible choice.
The numbers
| Grok 4.6 | Grok 4.7 | |
|---|---|---|
| Parameters | about 1.5 trillion | 2.1 trillion |
| Scale vs 4.6 | baseline | about 40% larger |
| Training data | previous mix | includes SpaceX engineering data |
| Real-time knowledge | yes | yes, with better recall on recent context |
| Vision and documents | yes | yes, with stronger document reading |
| Inline image generation | yes | yes, with more controllable edits |
The parameter figure is the one that gets quoted, and the arithmetic is simple: if Grok 4.7 is 2.1 trillion parameters and around 40% larger than its predecessor, Grok 4.6 sits near 1.5 trillion.
What the extra scale buys you
Reasoning that survives length. Grok 4.6 could follow a multi-step problem, but long chains of code, math, or research were where answers started to drift. Grok 4.7 holds the thread further before it needs to be nudged.
Better grounding on recent events. Both models have real-time knowledge from X. In practice, Grok 4.7 is more reliable at pulling the recent context into the answer rather than simply acknowledging it.
Documents and images. Upload a screenshot, a chart, or a PDF and the model answers from what is inside it. Grok 4.7 reads denser documents more accurately - tables and scanned pages included.
Image edits inside the conversation. Ask for a change to an image you just generated - a different background, a cleaner composition - and Grok 4.7 applies it without losing the rest of the picture.
When Grok 4.6 is still fine
Not every task needs the frontier model. Quick rewrites, straightforward summaries, and short factual questions are indistinguishable between the two. If you already have a workflow that produces good output on Grok 4.6, there is no urgency to change it.
The upgrade pays off where the task is hard: long reasoning chains, current events, dense documents, and iterative image work.
Which one should you use?
On grok47.lol the answer is simple - every conversation runs on Grok 4.7. If you want the newest model with real-time knowledge, vision, and inline image generation, you are already on it once you pick a plan.
Where to go next
- Grok 4.7 release date - what shipped with the release and how the rollout works.
- How to use Grok 4.7 - a walkthrough from sign-in to your first answer.
- Plans and credits - Starter, Pro, and Premium, billed monthly or yearly.
- The docs - setup and product detail.