The Three Charges Behind the Cancellation
Unlike the usual rage-quit post, the author lays out evidence like a bug report: at the same subscription price, the perceived usable quota keeps tightening, and heavy daily use hits the wall; the model's performance has unexplained swings, with output on similar tasks good one time and bad the next, suggesting silent downgrades or routing behind the scenes; and when problems arise, support goes silent beyond an autoreply. Any one of these could be tolerated, but with all three stacked, he chose to vote with his feet. The long line of "+1"s in the comments shows this isn't an isolated case.
A Common Ailment of Subscription-Based AI
Zoomed out, it pokes at the structural problem of all subscription-based AI: vendors sell a service with wildly fluctuating costs at a fixed monthly fee, and when compute is tight, the handiest lever is quietly tightening quotas and lowering the service tier—while the user faces an unauditable black box, where even proving "it got worse" is hard, and trust gets ground away in this information asymmetry. For anyone heavily dependent on a single subscription, it never hurts to keep an escape route to an API or a rival on hand; and for vendors, cancellation posts like this that itemize the evidence are actually an early warning they get for free.
via: Hacker News