GitHub Copilot is moving every plan to usage-based billing starting June 1st, and the internet is aghast - I've seen about 700 complaints on HackerNews. Well, GitHub is right to do it, and they are teaching every other AI-product company in public what billing v2 looks like in 2026.

A $10/month Copilot Pro plan, until a couple of weeks ago, gave you 100 "premium requests" against Claude Opus. One night of agentic coding could spend about $500 of API tokens for those aforementioned $10.

That was never a price, it was Microsoft eating the gap between what they sold (a flat fee) and what they delivered (variable inference cost on someone else's frontier model). That's marketing for you!

The unit of value moved underneath them

Two years ago, a Copilot request was a simple chat message so when you typed something the model typed something back. The cost was small, the answer was small.

Today, a Copilot request can be an agent that runs for four hours, spawns sub-agents, edits 80 files, and consumes 100 million tokens. That is the same SKU but we can agree it is not the same product.

GitHub kept the SKU in theory and absorbed the spread. There were users prompting "plan this out, don't stop, don't ask questions, make no mistakes, pls thx" and pulling several dollars of Opus output for one premium request. That was the loophole. The math was the loophole.

When your unit of value changes from under you, you have two options: Eat the spread until it eats you, or change the unit.

We understand that GitHub is changing the unit (a token), the meter (AI Credits, denominated in US cents), and the rate card (per-model multipliers) all at once.

That is the smart thing to do, and the token is the right primitive. Inference is the actual cost. Anything you bolt on top of it (e.g., request counts, message counts, "premium" tiers) is an abstraction that decays the moment the model behind it gets better at agentic work.

The execution, however….

Read GitHub's blog post and you will find the line "Plan prices aren't changing." That is technically true and substantively misleading. The annual Copilot Pro plan still costs $10/month. The model multipliers tell the rest of the story.

(Annual subscribers only. Monthly subscribers get pure token-metered billing instead)

This is like "your rent did not go up but we removed a room from your apartment".

Three execution choices that are going to bruise:

1. Annual subscribers got rugpulled. People who paid up front for 12 months of one experience are getting a different, materially worse one mid-contract. The legal exposure in Germany and Australia is real. The trust exposure is bigger and travels further.

2. The fallback model is gone gone gone. Today, when you exhaust premium requests, you fall back to a cheaper model and keep working. After June 1, you stop working until you top up. That is a feature deletion dressed as a billing change.

3. The narrative is.. uh… "Plan prices aren't changing" is the kind of sentence a CFO writes and a developer screenshots, it will follow this announcement for months and people will remember it..

The one thing GitHub is getting right is the preview bill rolling out in early May. Customers and admins can see what they would have been charged before the change goes live. That is the move every billing team should be stealing. People do not hate paying. People hate surprises.

This is what we mean by billing v1 → v2 at Solvimon

What the GitHub announcement actually is, under the marketing copy, is a billing platform update... Look at what they shipped at once:

• A new unit of measure (AI Credits, denominated in cents)

• Per-model rate cards with multipliers

• Pooled credits across an organization

• Budget controls at the enterprise, cost center, and user level

• Two pricing regimes running in parallel (annual versus monthly)

• A preview bill experience

• Prorated migration paths

• Three months of promotional included usage for Business and Enterprise

That is roughly six to nine months of billing engineering on a system already in production, serving a meaningful chunk of GitHub revenue. Every company building anything agentic is going to need a version of this. Most of them will discover, mid-pivot, that their billing v1 cannot bend this way.

Pricing is fashion so the rate card has to bend

The thing non-pricing people keep underestimating is just how often the rate card has to change. Vercel ships five or six pricing changes a month. Lovable changed pricing four times in 2025, while the frontier model providers quietly tighten their max plans on a rolling basis and manipulate it behind the scenes to control costs.

Pricing is fluid because the cost surface underneath it is fluid. When the model gets cheaper, you pass it on. When the model gets more expensive, you pass it on. When the model starts doing 100x more work per "request", you change the meter.

GitHub is doing the visible version of what every AI-product company is going to do in the next 18 months. Some will do it in one big June 1 release. Some will do it five times a month, quietly, because their billing infrastructure can carry the change. The second group is going to look a lot smarter to the market, and a lot less alarming to their customers.

Credits and tokens are here, that's a fact and that fight is over (even if I still don't like them as a customer). The argument is whether your billing system will let you keep moving once you do. Change unit, change rate, change pool, change region, change cohort, without a quarterly platform release every time.

That is the question every CFO at an AI-native company should be asking their CTO this quarter, while GitHub is still on the front page.