GitHub Copilot moved every paid plan to usage-based billing on June 1. Today is June 30. The first full month under the new regime just closed, and the screenshots are now arriving in everyone’s group chats.
The math, broadly. The old Pro tier was a flat $29 a month for unlimited completions. Pro+ was $50. Both are now token-metered through a credits system priced at API rates, with a monthly allotment baked in and overage billed at $0.01 per credit. If you used Copilot the way most people did in 2024, as inline tab-complete on a single file, the new system is roughly the same price or cheaper. If you used Copilot the way most people started using it in 2026, as an agentic coding loop that reads the whole repo, plans changes, edits across files, runs the test suite, and iterates until green, the new system bills you for every input token, every output token, and every cached token along the way.
Developers running that second pattern have been posting receipts. The most-shared numbers in the GitHub community thread and on Hacker News are bills jumping from $29 to roughly $750, and from $50 to roughly $3,000. One developer on Reddit clocked their highest-spend day at $187 for a single afternoon of agentic refactoring. The dominant emotion is a kind of stunned, retroactive math: people had been running thousand-token agentic sessions for fun, on the assumption that the meter was the same flat $29 it had always been.
GitHub’s defense is essentially the unsubsidized version of itself. Agentic coding burns through API tokens at a real rate. Microsoft was eating that cost at the old flat tier and the loss curve was visibly unsustainable. The new model passes API cost through. Anthropic’s Claude Code and Cursor’s $20 tier have been edging in the same direction for months, just less abruptly. The question for the rest of 2026 is whether the agentic coding category settles into “$200-500 per dev per month is normal” or whether teams quietly walk most of that workload back into the IDE and call it good.