AI Model Pricing Makes 0 Sense

AI Model Pricing Makes 0 Sense For the first time in two years, I'm actually considering switching over to ChatGPT instead of Claude for work. Claude isn't amazing, it is. It's just gotten so expensive in ways that don't make sense. And ChatGPT is also kind of amazing. So I just wanna talk about how pricing frontier models right now so you can make a smart decision for your business. The most important thing to understand paying for a subscription, your cost is heavily I mean ridiculously $20 subscription models will get you roughly $200 worth of usage. not enough usage for anyone who's actually using this for work. kicker, a $200 subscription to one of these models will get you the equivalent of $5,000 in usage. max plan, you're only paying 4% of what you would pay to pay for each token. Or another way to think of it, those tokens are marked up 25X compared to the subscription. that, if you're a serious models, you just have to be using a subscription. subscriptions were introduced when the were not concerned about revenue, they were just trying to get as many users as possible in the habit of using their model, and they succeeded, especially Anthropic. and they are compute-constrained and they are concerned with revenue. So naturally, they're doing everything they can to get us to pay for usage, Anthropic more than and they're also doing everything they can to lock you into their platform so that you can't use other models against your own data. Anthropic is making it so that if you're using something like OpenClaw, you can't use a subscription. You have to pay for usage. back Fable 5, which is easily the best model but you can only use it on a subscription through July 7th. After that, you have to pay for usage. A few days ago, I had Fable 5 help me refactor a code base. It was a single It ran for an hour. It spent five million If I had to pay for usage, that one prompt me more than $250. And honestly, given how capable it is and how big the task was, it probably it, but when you consider it actually cost me about $10 of my subscription, it's a tough And when you extrapolate that over our team of 25 people, we're difference between spending $5,000 a month and $125,000 month and $125,000 a month. So as far as companies like mine are concerned, come July 7th, we effectively no longer have access to Fable 5. Meanwhile, OpenAI just announced the ChatGPT 5.6 Sol which from what we know so far isn't quite as good as Fable 5. However, it's about half the cost per token, 30 per million compared to 50 per million and tends to spend far less tokens on the same task. And we're no longer comparing its 5, we're comparing it to Opus 4.8. Opus costs $25 per million output tokens so only slightly less expensive But again, Sol's also less tokens. So we'll have to see if OpenAI actually allows us to use Sol on the subscription, but if they do, it's just gonna be a much better model at a reduced cost would get from Claude. The lesson here is make sure you aren't stuck with a single model provider. knowledge base and your skills live somewhere like GitHub, between one model provider and another.