AI Quota Debacle - I Have Cancelled Everything.

I have cancelled all my subscriptions (£505.28 P/A  for All Products Pack and Ultimate AI) because I am so disgusted with the handling of the AI Quota.

I had never exceeded the Pro AI allocation (that I was now getting along with my All Products subscription) but in July I decided to get an Ultimate account as I was going to be using it a bit more, and didn't want to be keeping track of how much I was using.

July saw some pretty heavy usage, but in August it increased quite a bit. However, it was absolutely no problem and I still had plenty of allocation left at the end of the month (my allocation renewed on the 1st of the month)

Then September happened...and I burnt through my Pro **AND** Ultimate allocations in 3 days!!!

I raised a ticket, thinking there must be some mistake - but the reply was (as I am sure many of you are aware):

"The Quota Model has changed. That's how it is now. Suck it up".

I'm paraphrasing and have cut out the platitudes, but that is was what it amounted to.

I did a side by side test - both in PHPStorm - one using Junie and the other using Github Copilot plugin.

The task was to build a Tetris app in the browser.

The prompt was exactly the same, and both were set to use Claude Sonnet 4.

The results were very similar (I would say the Copilot one was slightly better and a bit quicker).

The costs,

Copilot: (easy to work out)
2 requests ($0.08). Immediately visible on Github.

Junie: (much more convoluted)
Remaining credits before task = 2,602,566.005
Remaining credits after task = 2,499,143.76
Difference =103,422.245 credits
Cost of credits = 3,000,000 top up credits costs $30 = 100,000 credits for $1
Total cost = (103,422.245/100,000) = $1.03!!!

This makes Junie 13 times more expensive than Copilot on a relatively simple task (it definitely works out even more on complex tasks once it starts looping the same answers/changes).

I reported this to JB, along with the prompt used and all the other details. Their only response was to ask for a screencast of both tests. I didn't record it and wasn't about to do it all again - they had all the details to repeat the test for themselves.

That was enough - I'm out!