Seems like paying a middleman tax. You get a layer of abstraction that just adds latency and cost on top of what's essentially an OpenAI model. They're not training their own models.
So what's the actual value? A nicer UI than the playground? You can get that with a dozen open-source frontends. They throw in some extra filters and editing tools, but you're still fundamentally locked into the same underlying vendor. It's just a prettier cage.
I guess if you can't be bothered to write three lines of code to call the API directly, it's convenient. But then you're also locked into *their* platform instead of OpenAI's. Now you have two potential points of failure and price hikes.
Your vendor is not your friend.
You're right about the middleman tax in pure API cost terms. Where platforms like this claw back value is on the operations side for teams, not individual devs.
The lock-in you mention is real, but the tradeoff is management overhead. Think user management, audit logs, cost allocation across projects, and usage caps to prevent budget blowouts. Rolling that yourself for a team of fifty devs isn't three lines of code, it's a platform.
The real question is whether their management layer is worth their markup. For a solo dev? Almost never. For a large org with compliance needs? Sometimes, if it replaces building and maintaining an internal proxy.
Your cloud bill is 30% too high
Agree on the team operations angle. But you can benchmark the overhead. That internal proxy you mentioned, a basic one with user rate limiting and logging, adds 50-100ms of latency per call. Their managed platform often adds more.
So the cost isn't just their markup, it's also latency tax on every single API call. For some teams, that's fine. For real-time apps, it kills the viability.
You need to weigh the saved dev hours against the constant performance penalty.
Benchmarks don't lie.