GitHub Copilot Can't Afford You Anymore

GitHub paused all new individual plan sign-ups and gutted the model lineup. The reason they gave is the quiet part said loud: agentic workflows consume far more resources than the plan structure can support.

Starting April 20, new sign-ups for Copilot Pro ($10/mo), Pro+ ($39/mo), and Student plans went dark. Existing subscribers keep access, but the model roster got slashed. Opus models were removed entirely from Pro. Opus 4.5 and 4.6 are being pulled from Pro+, leaving only Opus 4.7 for the highest-paying individual tier. Pro+ now gets 5x the token limits of Pro. Weekly consumption caps and session limits were added across the board. Business and Enterprise plans are unaffected.

On June 1, GitHub transitions to usage-based "flex" billing. Code completions and Next Edit stay free. Everything else, including Copilot Chat, CLI, cloud agents, Spaces, and Spark, consumes credits from a monthly allotment with pay-as-you-go top-ups. A new Max plan targets high-volume users.

This is the first concrete evidence that the economics of agentic coding workflows don't work at subscription scale. One Reddit user reported burning 62 million Opus 4.7 tokens in a single 24-hour session. The 2,694-upvote post about Claude having a bad morning suggests the compute pressure isn't just a GitHub problem.

We've been building habits around "AI is cheap, use it for everything." That era is ending, at least for frontier models at flat-rate pricing. Copilot is the canary. The move to usage-based billing is GitHub admitting that all-you-can-eat breaks when agents eat around the clock.

What to do: check which models your Copilot plan still supports. Budget for flex billing starting June 1. Seriously evaluate local models. The llama.cpp story in today's vibe coding section is relevant: local inference with Qwen3.6:27b hits 77.2% on SWE-Bench Verified. That's not Opus, but it's free after hardware costs.