Same here—I don’t know how the calculation works either. Before the change, I had only used it about a dozen times, but after it switched to 2000, it became over a hundred uses.
My usage suddenly shows that all 500 credits have been exhausted, yet I’m still able to use paid models normally. This appears to be a bug and makes me feel uneasy.
Interestingly, I conducted a test (one prompt) and observed that the resource consumption on the observation panel was the same as before. I’m a bit confused because I’ve seen different reports.
It seems to be an incomplete rollout, along with model adjustments, too. ~For example, although GPT-5.4 is still not available for grandfathered users, GPT-5.3 Codex high-reasoning has been replaced with medium-reasoning (no other GPT-5.3 Codex options).~
The usage log also doesn’t match the requests balance. I did a task with Opus 4.6 and another with GPT-5.3 Codex, which the logs shows totaled 3 requests. That made the usage bar jump by 12.
I’ve done some testing and I think it’s linked to the use of sub-agents. Each sub-agent uses up a request, so I reckon they’ve had to increase the quotas to take this into account.
If the usage pool was increased 4× and the actual cost of each request was increased 4×, then nothing has actually changed.
Maybe this will make way for lower cost promotions, since they previously couldn’t split a request credit in half. But I suspect it’s more likely that this will be a way to slip in partial cost increases. Like some models can now cost 25% more.
A bit worried that future new models will use more request times, while old models will use fewer request times, similar to a token-based billing method.
It seems like a bug, or is it distinguishing between subagent and main agent? Now one request consumes four requests (500 requests become requests, originally one request consumed 1, now it consumes 4? Don’t know how subagent requests are calculated)
I think this is that whole rollout thing. Before this happened, I was still on the old plan about 1–2 weeks prior, and once I exhausted my 500 requests/month, I enabled on-demand usage. For the past 2 months, the usage-based pricing was still per-request and kept increasing from $0.04 to $0.08 per request, which was still fair and transparent to me.
But then, color me surprised, around last Friday (or the Friday before, I forget), I was expecting the same $0.08 per request, then BOOM, after around 10 requests it spiked from $8 to $22 in usage. Wow.
I tried to create a support thread but got redirected to email instead, so I went that route and asked. Their reply was: “All the old pricing was changed to new pricing when the annual auto-subscription refreshed in February.” And that was it. The view still shows request-based usage, but the pricing silently switched to credit-based, which is very confusing and came without any prior notice.
For context, I did make a thread about this before and I think I misunderstood what they meant (English isn’t my first language), so I almost thought they were retaining the old 500-request plan. Had I known, I would’ve cancelled my subscription entirely at that point. Yes, other tools require a lot of setup, like a lot, but once I finished building the workflow that felt right for me, for example using OpenClaw and n8n, it gets pretty close to how I used Cursor, and with way more transparency on top of that.
Oh right, I almost forgot, about the on-demand usage blowing up to $12: I noticed that requests are capped at around 50–60K tokens each, meaning every time that limit is hit, or every tool call, it counts as one request. That adds up fast.
similar bug here. yesterday when finishing work, usage was exactly at 44. now when arriving to office it jumped to 176, so exactly *4.. seems like a ui calculation bug. only concerning thing for me is that it is 176/500 and not out of 2000..
