The pricing page says things like "$20 for 15M tokens" and "GPT-4o mini is $0.15/1M tokens". But the interface shows my usage in "messages". This is useless.
How many messages do I get per million tokens? Is there a simple conversion, or do they make it impossible to estimate my actual bill? I just want to know what I'm paying for before I hit a limit.
That's not a bug, it's a feature. OpenAI's entire business model relies on this exact opacity. If you could easily map "messages" to "tokens," you'd realize how wildly variable the cost of a single conversation can be. One "message" could be "ok" (a few tokens) or a pasted chapter of a novel (thousands of tokens). They're giving you a useless metric because a useful one would make you anxious about every lengthy prompt.
The incentive is to abstract away the real unit of cost so you think in conversations, not computational units. It's classic SaaS misdirection. You're not alone in finding it frustrating, but from their perspective, confusion is better than price shock.
But what about the edge case?
I see your point about the business incentive, but I think calling it a "classic SaaS misdirection" might be a step too far. The core issue isn't necessarily malice, it's that tokens are a technical unit most users shouldn't need to think about. The real failure, in my view, is that they're stuck between two audiences: power users who think in tokens and everyday users who think in conversations. Providing *only* a message count without any contextual translation tool is what creates the frustration.
They could solve this by offering a simple, conservative estimate in the interface, like "a typical message in this conversation uses ~X tokens." That would maintain the abstraction for most while answering the practical question for people trying to budget. Right now, they serve neither group well.
Reviews build trust.