June 1 marked a quiet turning point for anyone who writes software with help from a chatbot. With GitHub Copilot moving its remaining plans to usage-based billing on that date, the last of the major AI coding tools abandoned the flat monthly fee for unlimited help. The all-you-can-eat era is over, and a meter now runs in the background of most professional coding assistants.

How we got here

The shift has been building for a year. Cursor switched to a credit system in June 2025, Windsurf followed in March 2026, and Copilot's move on June 1 completed the set, as one industry roundup put it. Copilot now retires its old premium request counts in favour of GitHub AI Credits, billed by the token at each model's published rate.

The reason is unglamorous arithmetic. Reasoning models and agent-style workflows can swing the compute cost of a single request by orders of magnitude, depending on how long the model thinks and how many times it loops. A fixed price that covers a light user cannot also cover someone running agents around the clock. Providers were subsidising the heaviest users out of everyone else's subscription, and that math stopped working.

What it means for developers

A rough shape has settled across the market: a roughly $20 pro tier for everyday use and a roughly $200 power tier for heavy users, each bundling a monthly allowance of credits before per-token charges kick in. Claude Code, Cursor and Windsurf all land near those numbers, according to pricing comparisons published this spring. The practical effect is that cost now tracks usage, which rewards people who are deliberate about when they reach for an agent and adds up quickly for those who fire one off at every small task.

There is a culture shift buried in the pricing change. Treating model calls as effectively free encouraged a throwaway style, agents spawning agents and long speculative runs left to churn. Metering puts a number on all of it. Some developers will welcome the discipline. Others will feel the old magic, the sense of an assistant with no marginal cost, slipping away. The trend toward heavier agent use is not going anywhere, as our report on Cognition's billion-dollar raise showed. Either way, the bill now reflects the compute, and the compute was never really free.

Sources

  1. i. wilico.co.jp
  2. ii. www.developersdigest.tech
  3. iii. spectrumailab.com
  4. iv. www.nxcode.io

Commentarii · 0

Add · a · Comment