GitHub Copilot will switch to usage-based billing and AI credit system starting June 2026
GitHub Copilot will shift all plans to usage-based billing starting June 1, 2026. At that time, all premium request units will be replaced by GitHub AI credits. Each plan will offer a monthly allotment of these credits, and paid users can purchase more as needed.
The new system calculates usage based on token consumption, covering input, output, and cached tokens according to the API rates for each Copilot model. While this change aims to align costs more closely with actual usage, Copilot’s base plan prices will not change as a result of the transition.
To support the move, GitHub will release a preview bill experience in early May. This tool gives administrators and users advance visibility into their projected usage and associated costs ahead of the official switch. Following these updates, code completions and next edit suggestions will remain available in all plans without consuming AI credits. However, Copilot code review will now require both GitHub Actions minutes and AI credits.
GitHub says this transition is necessary because the previous model was unsustainable and did not capture the true computational cost of different use patterns. The company also expects usage-based billing to preserve service reliability while reducing the need to restrict high-activity users.
- AI Coding Assistant
- Freemium
- Proprietary
Related news
External links
- GitHub Copilot is moving to usage-based billingGitHub • Official source
- Microsoft to Shift GitHub Copilot to Token-Based PricingWindows Report
- Report: GitHub Copilot is moving to token-based billing from JuneNeowin
- Microsoft Brings Token-Based Billing to GitHub CopilotThurrott
- GitHub Copilot Moves to Token-Based BillingLet's Data Science
Comments
“this transition is necessary because the previous model was unsustainable” sheesh, i wonder why (sarcasm)
get worse at many complex tasks than before Have you actually used any of them for anything complex? Because ignoring anecdotal examples, the models have inproved enormously over the last year, since I started using them professionaly in day-to-day job tasks. But I agree the pricing is tough. You can actually burn the $20 in a couple of complex prompts over a few minutes, not hours.
It was a nice run while it lasted, but one of the key things will be how they start charging for usage after they switch over, because their models are still gimped compared to the official ones. So, if they go charging full official rates for them, there'll be absolutely no reason to use them, because the key reason to use Copilot was because you could get the thinking capacity of Frontier models for better prices, because Microsoft was tuning them down to be cheaper to run. Because You certainly didn't pay for the power of their tools vs your other options.
Vibe coding worked on 2 promises: inference cost will be reduced, and models will burn less tokens generating (well, trying and failing) code.
Microsoft absorbed the cost of theses promises for many years, but since models are more and more expensive (just as GPUs) and get worse at many complex tasks than before, people have to try harder to get some work done, and could reach $20 budget in just few hours if theses subscriptions were billed at the real cost (even some users reported burning some $100 using a $20 subscription).
So yes, GitHub Copilot, like all Copilot products and co, will cost and suck much more (even $200 subscriptions would still hide the real cost) because none of theses promises will ever be fulfilled by theses hungry companies that are still thinking that more people will pay more to code less.