The move, which was flagged by GitHub chief executive Thomas Dohmke in April, said the AI equivalent of telco-style bill shock to its Copilot service. Anyone using the fancier tools under the Copilot Pro, Pro+, Business or Enterprise plans is now looking at a quota of so-called “premium requests” before they’re pushed into metered billing.

These premium requests include anything requiring significant GPU grunt like Copilot Chat, Agent mode, Copilot code review, and Copilot Spaces. Each interaction, whether a prompt or a response, counts towards your monthly cap, and not all models are equal.

If you’ve picked GPT-4.5, prepare for your credits to evaporate faster than a crypto wallet at a Nigerian investment seminar. Its 50x multiplier means you’ll churn through your allowance 200 times quicker than if you stuck with Google Gemini 2.0 Flash, which barely sips credits at 0.25x.

Plans now come with fixed monthly allowances: Pro users get 300, Pro+ scores 1,500, Business plans offer 300 per user and Enterprise hands out 1,000 per user. Once that’s gone, you can opt into a metered rate of $0.04 per request, which makes this the cloud computing racket’s latest spin on the classic upsell.

Not surprisingly, punters aren’t happy with the move. A GitHub community thread discussing the shift has seen more than two dozen users slam the caps, claiming the move feels like a dark pattern to nudge upgrades.

As one user put it, “Feels like this is meant to push us into higher tiers instead of offering real value.” Others echoed similar frustration, suggesting that the changes punish those who actually use the product for its advertised strengths.

So GitHub just turned its AI assistant into a coin-operated coder.