On Monday, GitHub announced that starting June 1, every Copilot Pro and Pro Plus subscriber will be migrated from request-based billing to a token-based pricing model the company is calling GitHub AI Credits.

The official announcement landed on the GitHub blog and in community discussion thread 192948 on April 28. By Tuesday it was the loudest topic on Hacker News, the GitHub community forum and developer Reddit. The FAQ thread that followed, discussion 192963, has accumulated thousands of comments in 72 hours.

The mechanic is simple. Today, a Copilot Pro user pays $10 a month for a fixed allowance of premium requests. Starting June 1, that flat allowance disappears. Users will instead consume AI Credits priced according to the model selected and the number of input and output tokens. The token rates match GitHub's published API rates per model.

The Translation

This is GitHub admitting that flat-rate AI is no longer a viable business model. Copilot has been operating at a loss for months on heavy users. The Wall Street Journal and The Information have both reported in the last six months that Microsoft was subsidizing Copilot at roughly negative twenty dollars a user per month on the most active accounts.

Cursor moved to usage-based pricing in late 2025. Replit did the same. Anthropic's Claude Code agent has been credit-metered since launch. GitHub was the last major coding tool holding the flat-rate line. That line is now gone.

Pricing transparency cuts both ways. Light users on Pro will pay less than $10 a month. Heavy users running agentic workflows or long-context completions on Claude Opus and GPT-5 will pay more, in some scenarios significantly more. Annual subscribers get to keep the old plan until renewal, which is a polite way of saying GitHub is grandfathering its highest-value cohort and forcing the rest into the new model.

The Real Story Is the Token Cost

GitHub published the per-model rates in the same announcement. They mirror Anthropic and OpenAI's published API rates. So the marginal cost of a Copilot completion is now visible. That is new.

Developers who have done the math are already seeing the implication. Claude Sonnet at GitHub's published rate is roughly six to eight times more expensive per output token than GPT-4.1. Claude Opus runs higher again. The model picker inside Copilot now functions as a cost picker.

This is the same dynamic that hit Cursor users in late 2025. Power users started running default model selection like a budget allocation. Some teams banned Opus inside their internal Copilot config. Others flipped the other way and put Opus on every agent task because the productivity delta justified the spend. There is no consensus answer yet.

Microsoft's AI Margin Problem

Satya Nadella has said on the last three earnings calls that AI margin compression is a temporary effect. The Copilot pricing change is the first hard public evidence that Microsoft is no longer willing to absorb that compression at the developer-tool layer. Azure can still run AI workloads at sub-margin to hold market share against AWS Bedrock and Google Vertex. Copilot apparently cannot.

This also reveals the other end of the inference cost curve. Microsoft's Q3 capex was 25 billion dollars in memory alone, per the company's earnings call this week. That cost has to land somewhere. It is now landing in the per-token rate developers see at the bottom of their Copilot dashboard.

Three Watch Items

First, Copilot churn in May. If a non-trivial percentage of paying users cancel before the June 1 forced migration, the move is more painful than GitHub publicly anticipated.

Second, Cursor and Replit's pricing response. If GitHub's published per-model rates are lower than Cursor's, expect a pricing war by midsummer. If higher, expect Cursor and Replit to raise.

Third, enterprise Copilot. The change announced this week applies to individual Pro and Pro Plus plans. Copilot Business and Copilot Enterprise are still on seat-based pricing. The next pricing memo from Microsoft on the enterprise tier is the one that actually moves the AI coding market. Watch for that announcement before GitHub Universe in October.