GitHub just revealed how much AI models actually cost to run. And it changes how you should think about AI budgets.

From 1 June, GitHub Copilot is moving from flat-fee billing to usage-based pricing. Instead of paying a fixed monthly rate and using whatever you like, every interaction now consumes tokens, and every token has a price that depends on which model you choose.

**That makes the economics much more visible. **

Claude Opus 4.6 jumps from 3x to 27x. GPT-5.4 goes from 1x to 6x. That does not mean those models are 27- or 6-times better. It means they are dramatically more expensive to use. See the infographic for an overview.

This is a pricing signal, not a capability signal. A low multiplier means a cheaper model and more room to use it freely. A high multiplier means heavier compute and faster credit burn. Pick the wrong model for a routine task, and your allowance disappears a lot faster than you expect.

GitHub has also paused new individual plan sign-ups and tightened weekly usage limits, citing the compute demands of agentic coding sessions. Anthropic has also implemented reduced usage.

This is not just a GitHub story. It is a preview of where all AI pricing is heading.

Right now, most AI products still feel like flat-fee magic. Pay the subscription, use the tool, don’t think too hard about the cost underneath. But the economics have always been usage-based. Every prompt consumes compute. Every response consumes tokens. Every model has a different cost per unit of work.

GitHub is simply making that visible earlier and more explicitly than most.

If your business is budgeting for AI as though it were a fixed monthly software cost, this is the moment to revisit that assumption.