AI Pulse · Business Models & Winners

GitHub finished moving Copilot to a token based billing model on June 1. Basic code completions are still free, but agent mode and the premium models now pull from a monthly pool of “AI Credits,” and once that pool runs out you pay $0.04 per request. That’s the mechanics. The bigger issue is timing.

Since June, most teams haven’t felt any of this because GitHub handed out promotional credits worth $30 to $70 per user per month. Those promotional credits expire in September. That’s the month your team sees the actual baseline cost for the first time, not the discounted one you budgeted around all summer. If you approved Copilot based on what your invoice looked like in June or July, you approved a number that was never going to last.

There’s a second catch buried in the enterprise tier. GitHub advertises Enterprise seats at $39. To actually get that tier, you need GitHub Enterprise Cloud, which runs $21 a seat on its own. So the real price is $60 a seat, not $39. If you’re the one who signed off on the $39 figure, you’re the one who has to explain the $60 figure to whoever controls the budget.

None of this means Copilot is a bad tool. It means the pricing was structured to look cheaper than it is for the first few months, and September is when that gap closes. If you run a team on Copilot, go check your actual usage against the credit pool now, before the promotional credits vanish and you’re negotiating a renewal you didn’t see coming. This kind of gap between advertised and actual cost is exactly the sort of thing worth tracking automatically, which is the kind of thing we build into an AI command centre at our resources page. Treat this as a pricing pattern to watch for elsewhere too, not something specific to GitHub.