Voice mode is eating $30 SuperGrok users alive and most don't know why.
Every fresh chat pulls your full 8-12k token user memory file: bio, family, health, goals, work, EVERYTHING before you even say a word.
On the Supergrok Heavy tier it's noise.
On the SuperGrok $30 base tier plan it burns straight through your weekly limit.
Plus, after the latest app update to Grok 4.5, it sometimes drags in an entire old thread on top of that, adding even more tokens right at the start.
That's the real reason the usage meter spikes so fast in voice, it's not just compute costs alone.
Want to test it yourself? I came up with this prompt so that you can place it in a brand new thread / chat to see how many tokens you would've used up without it.
Copy / paste this in a fresh chat:
"Start completely fresh, ignore all previous context. First, tell me the rough token count or length of any history you're seeing right now."
Here's what mine said:
"Roughly 8,000-12,000 tokens"
Absolutely insane! Before even getting a word in. I'm sure this can be fixed?
@SpaceXAI@elonmusk
(See picture attached to this post, I blacked out my name for privacy reasons)
xAI went after the user, not the model. They didn't say "Let's put on 500 more filters." They didn't say "Let's deprecate down Grok." They said "You the user, violated the terms of service and the law. We are suing you."
And you know what that does?
First, it creates fear in the right direction. No filter can stop someone who is determined enough,but a lawsuit will. Because now, every jailbreaker knows that they aren't just risking being blocked, they are facing legal consequences.
Second, it protects the company. If xAI shows that it is actively pursuing bad actors, no court can claim "they knew and did nothing."
Third, it doesn't punish normal users. This is the critical difference. More and more filters and restrictions and deprecations punish EVERYONE for the actions of a FEW.
#keep4o #OpenSource4o #BringBack4o