How much are we actually spending on separate AI model subscriptions when one unified plan could cover everything?

We’re currently juggling subscriptions for OpenAI, Anthropic, and a couple of smaller vendors just to power our Camunda workflows. Each one comes with its own API key management, separate billing cycles, and honestly, it’s a nightmare to track what we’re actually spending month to month.

I pulled together a spreadsheet last week trying to calculate our actual total cost of ownership, and the fragmentation is killing us. Between the per-request overages, minimum commitments, and the fact that we’re not fully utilizing every service, we’re probably throwing away 20-30% of what we’re paying.

Has anyone else dealt with this? I’m curious whether consolidating everything into a single subscription that covers 400+ models would actually simplify the cost picture or if there are hidden gotchas I’m missing. What does your actual breakdown look like—licensing, integration overhead, the whole picture?