AI coding tools stopped pretending to be subscriptions on June 1 this year. That was the day GitHub moved every Copilot plan to usage-based billing, replacing its premium request units with GitHub AI Credits that are consumed according to the tokens an interaction burns, priced at the listed API rates for each model. A token, for anyone outside the machinery, is a chunk of text the model reads or writes; the more the model works, the more tokens it consumes, and now, the more you pay.
Code completions and Next Edit Suggestions stay unlimited, but everything else is metered. Annual plans are being retired altogether. GitHub was unusually frank about why. “GitHub Copilot simply is not the same product it was a year ago–it now powers far more complex, agentic workflows that consume far more compute,” the company wrote, describing the change as aligning pricing to actual usage and costs.
That sentence is the whole story. A subscription works when usage has a natural ceiling: one human, typing, eight hours a day. An agent has no such ceiling. Set a goal, and it will plan, write, run tests, read the failures and try again, for hours, consuming compute the entire time. Flat pricing meant the heaviest agent users were being subsidised by everyone else, and the vendors have decided everyone else has subsidised enough.
What AI coding tools cost now
The numbers are concrete. One AI Credit equals US$0.01, and the included monthly allowances run from 1,500 credits on Copilot Pro to 7,000 on Pro+ and 20,000 on Max, with Business and Enterprise users drawing 1,900 and 3,900 credits from pooled organisational allowances. A preview bill experience launched in May so users and admins could see projected costs before the switch, and the June 1 release added user-level budget controls.
The transition has not been smooth. Reports in Ars Technica, TechCrunch and The Register described developer backlash, with users posting screenshots projecting overage bills running from hundreds to thousands of dollars after burning through allotments quickly. The mitigating detail buried in the noise: overages only apply if a user sets an additional spending budget, and leaving it at zero stops Copilot rather than charging beyond the subscription.
Your tool dies before your card does, which is either a safety net or a productivity cliff depending on the day of the month. GitHub is not an outlier; the whole category repriced within weeks. Cursor, Windsurf/Devin, and the Anthropic API have all shifted pricing this month, and the most capable models are getting dearer rather than cheaper: Anthropic’s new Claude Fable 5 lists at US$10 per million input tokens and US$50 per million output tokens, twice the rate of Opus 4.8, with subscribers given free access until June 22 before it starts drawing on usage credits.
Inside Copilot, Fable 5 went generally available on June 9 with a catch: enterprise admins should read twice: it requires data retention of up to 30 days for Anthropic’s safety classifiers, and the policy defaults to off until an administrator enables it.
The cloud bill has come for the dev team
Engineering managers have seen this movie. It is the cloud billing model arriving at the dev toolchain, roughly two decades after it arrived at infrastructure, and it brings the same consequences. Budgets become a daily concern rather than an annual procurement line. Someone on the team becomes the person who watches the dashboard.
The cheap-model-versus-frontier-model choice turns into a cost decision made per task, because a quick chat and a multi-hour agent run no longer cost the same. There is a defensible logic to all of it. Metered pricing is honest pricing, and the alternative was vendors throttling heavy users behind opaque rate limits, which is what much of the past year’s friction quietly was.
But the change landed hardest on the developers who embraced agentic workflows most enthusiastically, the very behaviour every one of these vendors spent two years encouraging.
The pitch was: let the agent run. The bill now says: within budget.
The honest advice for teams this month is unglamorous. Check who actually has budgets set, watch the first full billing cycle closely, and treat model selection like instance selection. The era of not thinking about what your AI coding tools cost per task lasted about three years. It is over now.
(Photo by GitHub)
See also:OpenAI building GitHub alternative for developer toolchains
Want to dive deeper into the tools and frameworks shaping modern development? Check out the AI & Big Data Expo, taking place in Amsterdam, California, and London. Explore cutting-edge sessions on machine learning, data pipelines, and next-gen AI applications. The event is part of TechEx and co-located with other leading technology events. Click here for more information.
Developer Tech News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.