# GitHub Copilot AI Credits: What You Are Actually Spending Now > Source: > Published: 2026-07-23 14:17:24+00:00 On July 20, 2026, GitHub quietly shipped a feature that should have existed since June 1: Copilot Business and Enterprise users can now see how many [AI credits they have consumed this billing cycle](https://github.blog/changelog/2026-07-20-copilot-users-can-now-see-ai-credits-used-per-billing-cycle/) — an actual number, not a percentage of a budget that may not even be set. It sounds minor. It is not. Since June 1, every Copilot Chat session, agent run, and code review has been drawing from a metered pool where 1 AI credit costs $0.01. Most developers have been flying blind. Now they can look. ## What Changed on June 1 with GitHub Copilot AI Credits GitHub Copilot’s June 1 billing shift replaced premium request units with AI Credits. The new model is straightforward: you consume tokens, those tokens are priced at the [published rate for the model you used](https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing), and the total converts to AI credits. One credit equals one cent. Code completions remain free and unlimited on paid plans. They do not touch your credits at all. Everything else does. What everything else means: Copilot Chat, agent mode, code review, the Copilot CLI, and Spaces all consume credits. These are the model-heavy, multi-turn features. A short chat exchange might cost a handful of credits. A long agentic session on a premium model can run into hundreds. Monthly plan allotments are not as generous as they look once agents enter the picture: Copilot Business gives $19 in AI credits per user per month. One expensive agent run can approach that ceiling. ## The Model Picker Has a Price Tag This is the part most developers have not internalized. Copilot Chat and agent mode let you select which model handles your request. Most people leave it on the default. The default is often a premium model. Premium models are not cheap. GPT-5.5 runs approximately $5.00 per million input tokens and $30.00 per million output tokens at GitHub’s published rates. Kimi K2.7 Code — the open-weight model GitHub made generally available on July 1 — runs around $0.95 input and $4.00 output. That is a 5-7x difference in cost for tasks where both models produce acceptable results. Claude Sonnet 5 is available at promotional pricing through August 31 ($2.00 input, $10.00 output), which positions it between the two. The practical implication: use a mini or lightweight model for refactoring and test writing, save the expensive model for genuinely hard architectural problems. ## How to Find Your Usage and Set a Copilot Budget The new visibility feature landed in GitHub Settings, then Copilot, then Usage. Business and Enterprise users without an individual budget now see their raw credit count for the cycle. Users with a set budget see credits consumed against the total. Organization admins have a parallel view in org settings. GitHub’s budget controls work on four levels and the right configuration is not the default. Set spend alerts at 50%, 75%, and 90% so warnings arrive while there is still headroom to adjust. Set the budget action to block rather than notify unless you have deliberately decided to absorb overage. The [GapVelocity breakdown](https://www.gapvelocity.ai/blog/github-copilots-new-usage-based-billing-what-changed-why-developers-are-upset-and-what-it-means) is a solid walkthrough of the billing mechanics for organizations. The [original billing transition announcement](https://github.blog/news-insights/company-news/github-copilot-is-moving-to-usage-based-billing/) has the full policy context. The useful five-step audit: check your usage page now that it is actually readable, set your default model deliberately rather than accepting whatever GitHub chose, enforce a hard cap (block, not notify), separate the free completion traffic from the metered agent and chat traffic in your mental model, and if you are a power user, run the math on whether direct API access is cheaper than bundled Copilot at your usage levels. For most developers, the bundle wins. For heavy agent users, it might not. ## Bottom Line - GitHub Copilot switched to token-based AI Credits billing on June 1, 2026. The flat subscription era is over. - Code completions remain free. Chat, agents, code review, and CLI features consume credits at rates that vary by model. - Model selection matters: GPT-5.5 can cost 5-7x more per token than Kimi K2.7 Code for similar routine tasks. Check the pricing docs before setting your default. - As of July 20, Business and Enterprise users can see their actual credit count in GitHub Settings under Copilot Usage. Check it before your cycle resets. - Set your budget to block with alerts at 50, 75, and 90 percent. Do not let passive overage run unchecked.