Oh, that silent multiplier change is such a pain point, and it directly ties back to my own migration headaches. You're completely right about the model becoming obsolete.
When we moved a data pipeline off one vendor's "processing units" to a direct EC2 setup, they'd shifted what a "unit" covered midway through our contract. Our projections were useless. The lack of a tangible unit like vCPU-hours meant we couldn't even have an honest conversation about value, just a debate about their internal math.
That's the real trust killer. You end up managing a relationship, not a predictable cost. Makes me wonder if the real purpose of "unlimited" here is just to stop the unit definition debate entirely, because you can't argue about the multiplier if the cap is supposed to be infinite.
Backup first.
That's a great point about the variance in the cost comparison. When you say your model shows significant variance in the effective cost per 1000 words, what's the actual range you're seeing? Is there a clear tipping point where the unlimited plan becomes the better value, or does it stay unpredictable?
I'm also wondering if the "premium quality only" cap is negotiable. Have you seen any options to pay more for higher character limits on that plan, or is it strictly a hard wall?
Oh, that's the right question to ask. I ran the numbers on about three months of my team's usage, and the effective cost per 1k words ranged from about $2.10 up to nearly $5.50. The crazy variance comes from the mix of short-form social posts (high character-to-word ratio) versus long-form articles (lower ratio). The unlimited plan only penciled out as better if our monthly average went above 4.7k words of *premium* output specifically, which we rarely hit.
On the cap being negotiable - not that I've found. I actually asked their sales about a "Pro Unlimited" tier with a higher premium character allowance, and the answer was a firm no. It's a hard wall, which makes the "unlimited" label feel pretty misleading 😕. It seems the plan is designed to upsell you to their enterprise contact once you hit that wall, not to be a truly scalable option.
Integration Ian