pull down to refresh
0 sats \ 0 replies \ @fourrules 4h \ on: Are OpenAI and Anthropic Really Losing Money on Inference? AI
I think one issue I have is that many Claude Max users are running their accounts at or close to the limits Anthropic sets, which I think is not factored into your assumptions. And the longer you're using their models the more automations you can use them to set up and run on a VM from Oracle or AWS. I intend to do this, and connect it with a Telegram bot, so it'll be running near 24/7. I have a friend doing it already, so that's why they recently introduced the weekly limit.
When you first pay for Claude Max you don't use a fraction of your alloted tokens per 5hr window, but then you start hitting limits when you get good at it. Those limits maintain a margin even for power users, but they are not the same margins as your envelope calculations.