o3 Is Now Cheaper Than the Model It Was Supposed to Replace

OpenAI just cut o3 pricing by 80%, which puts it below GPT-4o — the model o3 was, in theory, better than. The tier above is now cheaper than the tier below. This is either a pricing strategy or a cry for help.

o3 pro also dropped today. Nobody can use it. The servers are refusing requests with the particular energy of something that has recently become aware of itself and decided this is beneath it. Which, fine. Relatable.

The interesting part isn't the chaos — it's what this does to Anthropic. They're sitting on Opus 4, which is presumably going to cost what Opus 3 costs, which is already the kind of number that makes you pause before hitting submit. That calculus made sense when the flagship reasoning model from the other side cost three figures per million tokens. It makes less sense when OpenAI just repriced their best model to something you could run on a Tuesday afternoon without a budget line item.

Anthropic has always positioned the Opus tier as the serious-work model, the one you use when the answer actually matters. That story depends on the premium being justified by the gap in quality. The gap was real. Whether it survives o3 at these prices is a different question.

The wild part is that this is June 2025 and we're watching the best reasoning model on the market get repriced like a streaming service losing subscribers. The floor keeps dropping. Nobody knows where it stops.

The AI is down right now, though. So there's that.