Da7em@Da7_Tech
What a disaster.
This genuinely goes against everything people expect from the values behind open-source models.
Yesterday, Zcode was released — ZAI’s own app, built to work with their GLM models.
I tried it on the Pro plan, the same Pro plan they describe as “15x Claude Code.”
One single task consumed 60 million tokens from my limits in less than an hour, burning through the entire 5-hour limit without even completing the task.
And the worst part?
Their own app clearly showed the task used under 2 million tokens.
But according to my tests, ZAI still deducted around 60 million tokens from both the 5-hour limit and the weekly limit.
That strongly suggests cached tokens are being counted against me as real usage.
So in practice, I am not even getting 5% of the value of my subscription.
The same task was completed by Codex and MiniMax on their $20 plans.
That says a lot about ZAI’s current infrastructure.
I really hope they fix this, because I do want their models to succeed. I believed in the project enough to subscribe for a full year.
But so far, for more than a month, I have not been able to actually benefit from my subscription.
Let’s see what happens in the next few days.
