Key findings
- Copilot Business costs US$19 a month and adds 1,900 AI credits per seat to the billing entity's pool. Enterprise costs US$39 and adds 3,900. At US$0.01 per credit, included usage equals the seat price.
- Organizations and enterprises allow paid usage beyond the pool by default. Spending limits send alerts until you enable Stop usage when budget limit is reached.
- Per million output tokens, GPT-6 Luna costs US$0.50, Claude Opus 5.5 US$20, and GPT-6 Astra US$50, a 100-fold spread from cheapest to most expensive.
- A code review costs an estimated US$0.05 to US$1 of credits at Lite effort and US$0.25 to US$5 at Balanced, plus GitHub Actions minutes. GitHub picks the review model itself and keeps it private, so budget from those ranges.
- Of 17 public accounts that assess billing between October 22, 2025 and July 2, 2026, 9 call it unpredictable and 8 workable. Of the 9 published after the April 27, 2026 announcement, 6 call it unpredictable.
Budget for Copilot as metered spend. Since GitHub moved every plan to usage-based billing in June 2026, a seat buys a share of a pooled AI credit allowance. Chat, agents, Copilot CLI, and code review draw on it by model and token count. Completions stay unlimited. Three settings now control the bill: spending limits, model policy, and who pays for code review.
What changed
Copilot previously counted premium requests. Chat and agent requests drew down a monthly allowance with model-specific multipliers. Usage-based billing meters input and output tokens, including cached tokens. GitHub prices them by model and converts the charge to AI credits at US$0.01 each.
Completions and next edit suggestions stay unlimited on paid plans. Everything else that calls a model draws credits, and GitHub lists what that covers.
- Copilot Chat and Copilot CLI.
- The Copilot cloud agent and third-party coding agents.
- Copilot Spaces and Spark.
- Copilot code review, which also consumes GitHub Actions minutes for its agentic context gathering.
What a seat buys now
Organization seats fund a shared credit pool. GitHub's example gives 100 Copilot Business users 190,000 credits, so heavier users draw on unused credits from others. Mid-cycle license additions expand the pool immediately. Removals shrink it next cycle. Unused credits expire when the pool resets at 00:00 UTC on the first day of each month.
Individual plans carry their own allowances, each a base amount plus a flex allotment.
- Copilot Pro: US$10 per month, 1,500 total monthly credits.
- Copilot Pro+: US$39 per month, 7,000 total monthly credits.
- Copilot Max: US$100 per month, 20,000 total monthly credits.
- Copilot Business: US$19 per granted seat per month, 1,900 credits per user, pooled.
- Copilot Enterprise: US$39 per granted seat a month on GitHub Enterprise Cloud, adding 3,900 pooled credits per user.
Where the money goes
GitHub says a short chat question on a lightweight model can cost a fraction of a credit, while a long agent session across many files on a frontier model costs far more. Copilot does not switch to a cheaper model when a budget runs out.
The model price table cannot predict code review costs. GitHub's models and pricing page says it selects the review model automatically and keeps it private, so per-token costs vary between reviews. Its code review page adds that reviews can use models your organization has not enabled. Budget from GitHub's per-review ranges, and ask your GitHub account team to name the models if your model policy requires it.
Automatic reviews bill the pull request author by default. Reviews for unlicensed members bill the organization outside user budgets. Microsoft's 2026 command-line-agent rollout study warns that organization-wide token spending can reach millions of dollars annually.
What engineers reported
Of 65 public accounts GILD tracks for GitHub Copilot, 17 assess billing. Eight call it workable and nine unpredictable. Individual builders write most of both groups. Several chose Copilot for its monthly allowance, then lost models or faced higher multipliers in April 2026 before credits arrived.
Accounts from inside companies point both ways. One engineer reports that their whole department moved to Claude Code after the June change, when a few dollars of overage became a monthly budget exhausted every two or three days. Another says their company compared costs and kept Copilot for included usage at provider prices across model vendors. A third expects negative ROI and less AI use in a company standardized on Microsoft. A fourth reports that only about one in three people on nearby teams use Copilot much, which leaves the pooled credits to the heavy users.
What to decide before the next billing cycle
Budget controls address both outcomes in these accounts, a pool exhausted in days and seats nobody uses. The GitHub Copilot AI Credit Budget & Policy Checklist explains each control.
- Choose whether paid usage past the pool is allowed, and set an enterprise spending limit with Stop usage when budget limit is reached.
- Set a universal user-level budget above US$19 or US$39, with time-limited overrides for named power users.
- Assign users directly to cost centers and cap each unit's draw on the pool to what its licenses fund.
- Restrict frontier models to the organizations that need them, and decide who is billed for code review.
Methodology
GILD checked GitHub's documentation and plans page on October 12, 2026 for plan, credit, budget, and pricing facts, and rechecked the code review model statements on October 25, 2026. Practitioner counts come from 65 public accounts in GILD's GitHub Copilot Signals, opened from October 12 to October 26, 2026. Of those, 17 assess billing, all published between October 22, 2025 and July 2, 2026. 20 discuss usage cost, published between October 22, 2025 and September 17, 2026. The counts describe those accounts alone. GILD calculated the 100-fold model price spread from GitHub's published prices.
Products in this piece
Sources
- Official documentation
- 7
- Public source
- 6
- GitHub Copilot docs: Usage-based billing for organizations and enterprises
- GitHub Copilot docs: Plans
- GitHub Copilot plans & pricing
- GitHub Copilot docs: Models and pricing
- GitHub Copilot docs: Getting started with budget controls
- GitHub Copilot docs: Optimizing your budget configuration
- GitHub Copilot docs: Copilot code review
- Hacker News: GitHub Copilot is moving to usage-based billing (thread, Apr 27, 2026)
- Hacker News: Kon5ole on a department's budget after the June 2026 change (Jul 2, 2026)
- Hacker News: K3UL on a company cost comparison after June 2026 (Jul 2, 2026)
- Hacker News: koyote on ROI in a Microsoft shop (Apr 27, 2026)
- Hacker News: sandos on pooled credits across teams (Apr 28, 2026)
- Murphy-Hill, Butler, and Savelieva: Microsoft's early 2026 rollout of Claude Code and Copilot CLI (arXiv, Jul 1, 2026)