Copilot curbs coding capacity credits

Everyone saw this coming

One of the best values in AI computing for solo developers has been the $10/mo tier for GitHub Copilot, not to be confused with the Copilot being stuffed into various MS products for years. This tier gives users CLI access to a harness that has a wide variety, and ever-changing, list of 3rd party models you can use. The key part is they're billed as "premium requests", a monthly bucket that you get 300 of. And one of them can do a lot, like a full GPT-5.4 high code review. The context window isn't as big, but similar things would take a massive amount of capacity in either Claude Code or Codex itself.

2 weeks ago they pulled Opus models from the platform. Today it's announced that that's all going away in favor of what looks like direct API billing, not subsidized. The end of the blog post is the most telling — "The bottom line: Plan prices aren’t changing." is a bit like saying "the 2 liter containers of Coke you buy at the grocery store didn't change in price, they're just now 8 ounces. Also, our sweetener is a lot worse. You won't like it."

It remains to be seen how bad this is, but most likely, this will be the first one in my current workflow to go or to see changes in June. At least we have an extra month. It's also important to note that if this truly is the first monthly subscription to completely drop "hidden subsidies", i.e. all straight API billing, and it works, the big 2 will be sure to follow suit.