The End of the Free Lunch
GitHub is finally admitting what everyone in the AI industry already knew: subsidizing unlimited AI code generation for a flat monthly fee was a burning pile of money. Starting June 1, Copilot will switch to a usage based billing model, charging users by the token consumed. The company claims this aligns pricing with actual compute cost, but let’s call it what it is: a desperate attempt to stop hemorrhaging cash on power users who treat AI like an infinite vending machine. The old model where a trivial chat cost the same as a marathon autonomous coding session was a ticking time bomb.
The Agentic Parasite Problem
The real villain here is the rise of agentic AI assistants like Openclaw, which can burn through millions of tokens in a single session. Ed Zitron recently leaked internal GitHub documents showing week over week inference costs nearly doubling since January. GitHub says it can no longer absorb the escalating cost, but that’s PR speak for we built a business model that let superusers feast on our GPUs. Under the new system, you get a fixed pool of AI Credits. Go over? You pay API rates ranging from $4.50 to $30 per million output tokens depending on the model. Simple code completion is free, but code reviews? Those now eat your GitHub Actions minutes. It is a shrewd recalibration designed to punish the heaviest users while keeping casual developers hooked.
Source: Arstechnica
