GitHub Copilot Just Switched to Pay-Per-Token. That's a Bigger Deal Than It Sounds.

On June 1, GitHub Copilot did something that would have been unthinkable a year ago: it dropped flat-rate pricing and moved every plan to token-based billing. The tool that arguably normalized the $10-a-month AI subscription is now charging by usage, like a utility. It is worth pausing on what that signals, because Copilot is rarely an outlier — it is usually a bellwether.

What actually changed

Under the old model, you paid a fixed monthly fee and Copilot's costs were GitHub's problem. Heavy users were subsidized by light ones, and the price stayed flat no matter how much you leaned on it. The new model bills against actual token consumption across all plans. In plain terms: you now pay closer to what your usage really costs.

The official reasoning is the one every provider eventually arrives at. Reasoning models and agentic workflows burn far more tokens than a simple autocomplete ever did, and a single flat fee cannot absorb that variance forever. When the underlying cost is variable, the pricing tends to follow.

Why this is part of a pattern, not a one-off

Copilot is not alone, and that is the real story. Over the past year the gravity in AI pricing has shifted in one direction:

• Frontier reasoning models (the o-series, DeepSeek-R1 and their successors) made per-request costs wildly variable, which flat fees handle badly.
• Agentic tools that chain dozens of model calls per task made "unlimited for $20" quietly unsustainable.
• Efficiency gains pushed the price of a single ordinary message so low that, for most people, metered billing is now cheaper than a subscription — not more expensive.

Put those together and the flat monthly fee starts to look like a transitional artifact: a simple price for a simpler era. The market is converging on the same idea that infrastructure landed on decades ago — you pay for what you use.

The catch nobody mentions

There is a reason flat fees persisted so long despite the economics: they feel safe. "$20, no surprises" is psychologically easier than "however much you used." A metered bill asks you to trust that your real usage is modest — and for most people it genuinely is, but the uncertainty is real.

So the shift to pay-per-token only works well when two things are true:

1. You can see your spending as it accrues, not just at the end of the month.
2. You can cap it, so a runaway agent or a curious weekend cannot produce a shock.

Every provider dashboard now offers spending limits for exactly this reason. The honest advice, whether you are on Copilot's new pricing or anything else, is to set one before you start. It costs thirty seconds and removes the only real downside of metered billing.

What it means if you use AI for more than code

The Copilot change is specific to developers, but the logic is general. If pay-per-token is becoming the default way to buy coding AI, it is reasonable to expect the same drift in how we buy AI for writing, research, and everyday chat. The subscription bundles will not vanish — they still suit the heaviest daily users and anyone who relies on bundled extras like image generation or voice. But for the large middle of users whose usage is bursty, metered pricing increasingly wins on cost.

This is the same premise behind bring-your-own-key tools generally — including ByteChat, which runs on your own provider keys at pay-per-token cost rather than a flat fee. The point here is not the tool; it is that the broader market is now moving toward the model these tools were built around. When GitHub Copilot adopts your pricing philosophy, the philosophy has arguably gone mainstream.

The takeaway

Copilot's switch is a small headline with a large implication: flat-rate AI is starting to give way to metered AI, because the underlying costs demanded it. For most people that is good news — pay-per-token usually means a smaller bill — provided you do the one thing the new era rewards: set a spending cap, then stop worrying about it.

Frequently asked questions

Did GitHub Copilot switch to pay-per-token pricing?

Yes. As of June 1, 2026, GitHub Copilot moved every plan from a flat monthly subscription to token-based billing, charging against actual usage rather than a fixed fee.

Is pay-per-token cheaper than a flat AI subscription?

For most people it is. Flat subscriptions are priced for heavy daily users, so anyone with bursty or moderate usage tends to pay less under metered, pay-per-token billing. Heavy power users who lean on a tool all day may still come out ahead on a flat plan.

How do I avoid surprise bills with token-based AI pricing?

Set a spending cap at the provider before you start, and use a dashboard that shows spend as it accrues rather than only at month's end. Every major provider offers usage limits; turning one on takes about thirty seconds and removes the only real downside of metered billing.

Watching where AI pricing is headed? The trend lines all point the same way: pay for what you use.