Я нанял дешёвых ИИ в помощники GPT-6 Astra. Стало дороже 😢

After the last issue I didn't want to turn off Astro . She made the website, put together the game, and you already catch yourself wondering what else you could give her. But using it for every little thing is insanely expensive for me . In the API, a million input tokens without cache cost $10. A million

weekends - 50. I have pro. But if you have a plus, the question is also whether the limit is enough to complete the work. You've already paid for the month , but the next edit has to be postponed until the reset. Therefore, let's take a closer look at what these savings mean for the

subscription. That's why I want to keep Astra in operation, but delegate some of its tasks to cheaper models. Let Sol and Luna write the code, and Astra distribute the work and check the result. Although it may also be cheaper to manage than SO, Astruvat on the plan and final verification. We made four versions of one program. Let's compare

how they work, what mistakes they made, and how much the result cost. And a small spoiler: the option with cheap assistants ended up being almost three times more expensive than Astro without a team. Let's imagine a small workshop. She prints t-shirts, mugs and stickers. Orders come from different clients. Each has its own circulation and

term. We need to see what is already ready, what has been put into work, and where we are late. We order such a panel for our models. You can add an order, change it, drag it to another column. There is a search, filters and export to a table. After restarting the server, the data should remain.

You also need to be able to open an order and change the status on your phone. Everyone was given one project template, one layout, and 30 fictitious orders. There was no database or working buttons in the template yet . The libraries are pre-installed so you don't have to compare download speeds. Ara does everything herself at first. In the second