Opus 4.7 is here. It’s more expensive and it doesn’t even know that Opus 4.6 exists
A brief illustration of the challenges of constantly evolving AI: 1.) It doesn’t even know previous models of itself exist, and 2.) Understanding what you’ll pay is not easy.
Screenshot of me asking Claude Opus 4.7 how its usage costs compare to Claude Opus 4.6. Emphasis mine, obviously.
Today is a two-part example of where AI is still frustrating and, at times, maddeningly dumb in completely unexpected ways.
The concepts that are really important to understand if you use AI semi-regularly are:
AI isn’t knowledgeable in real-time on what it’s capable of when things change. You have to make it do research. The models are updating quickly. But the training data used to teach AI doesn’t update as fast. So there is a lag in what it literally knows about itself.
AI is getting more expensive but it is difficult to see directly. AI costs (tokens, credits, API calls, etc. etc. etc.) are rising. It is complicated to understand because the goal posts move and it’s tough to compare apples to apples. But it makes a difference and you shouldn’t always just default to the latest model.
The screenshots below illustrates both. Claude just released Opus 4.7. It is their most powerful model yet. I wanted to know what it would cost to use, because I do some pretty in depth work and if I’m not careful, I burn through credits and get put in AI “jail” (think of Monopoly where you either have to keep rolling the dice or pay the bank to get out of Jail).
I saw Opus 4.7 and asked Claude about the cost difference between Opus 4.6 and Opus 4.7. At first Claude didn’t even know Opus 4.6 existed. I’ve been using it for a while, yet Claude Opus 4.7 (the smartest model they have) told me confidently that there was only Opus 4.1. I’ve taught it to actually do current research, so it corrected itself, thankfully. But I was paying attention.
Claude Opus 4.7 casually telling me there’s no such thing as Claude Opus 4.6(!)
Then it told me all about the pricing. How this it’s the same, but not really, how it charges based on messages but it also uses more tokens in those messages so yeah, it’s more expensive. It’s all a bit exhausting to keep up with.
Why does this feel like I’m trying to follow which cup the ball is under?
So, Opus 4.7 is here. It’s (probably) very smart and better than previous models. It’s also more expensive, although it won’t exactly be obvious if you just look at the unit price.
Long story short: Don’t assume AI knows what’s it’s most current versions are capable of, or if they even exist. And use a model that does the job well enough for the specific task you’re working on. Haiku, Sonnet, and Opus are all very good. They consume very different amounts of tokens to deliver an answer to your question. So think through the complexity of the question. You don’t need to pay a doctor to tell you how to put on a band-aid.
