GitHub Copilot Billing — Move to AI Credits (June 2026)¶
What it is¶
Starting June 1, 2026, GitHub is replacing the existing "premium request" allowance model for Copilot Business and Copilot Enterprise with a usage-based billing model denominated in GitHub AI Credits. The transition was officially announced on the GitHub Blog and is documented in the GitHub Docs reference page; the change was reported in advance by tech press (Neowin / Ed Zitron) about a week earlier in late April 2026.
The change reflects an industry-wide shift: flat-rate "unlimited" or high-cap AI coding subscriptions at $20–$30/month have proven unsustainable as agentic workloads (multi-step coding agents, long sessions, frontier models) consume far more tokens than the original chat use case the pricing was modelled on. Anthropic moved enterprise customers to token-based billing first; GitHub is following suit for the organizational SKUs.
How it works¶
GitHub AI Credits¶
- 1 AI Credit = $0.01 USD (fixed conversion).
- Every Copilot interaction consumes input tokens, output tokens, and cached tokens. Token consumption is converted to AI credits using a per-model rate.
- Lightweight chat questions cost a fraction of a credit; long coding-agent sessions across multiple files with a frontier model cost substantially more.
What is billed in AI Credits¶
Billed:
- Copilot Chat
- Copilot CLI
- Copilot cloud agent (the asynchronous coding agent — see github-copilot-agents)
- Copilot Spaces
- Spark
- Third-party coding agents invoked through Copilot
Not billed (remain unlimited on all paid plans): - Code completions - Next edit suggestions
Included monthly credits¶
Standard included allowance per assigned license:
| Plan | AI credits per user per month | USD equivalent |
|---|---|---|
| Copilot Business | 1,900 | $19 |
| Copilot Enterprise | 3,900 | $39 |
These match the existing per-seat subscription price, so the included allowance is effectively "your subscription dollars back as AI credits".
Promotional period¶
For the first three months (June 1 – September 1, 2026), existing Copilot Business and Enterprise customers receive a higher promotional allowance to soften the transition:
| Plan | Promotional credits/user/month |
|---|---|
| Copilot Business | 3,000 |
| Copilot Enterprise | 7,000 |
After September 1, 2026, included usage returns to the standard 1,900 / 3,900 amounts.
Pooling¶
Included credits are pooled at the billing entity level, not per-user buckets. An enterprise with 100 Business seats gets a single shared pool of 190,000 credits per month. This means heavy users (engineers running long agent sessions) can draw from the slack created by lighter users. Adding seats mid-cycle increases the pool immediately; removing seats mid-cycle does not shrink the pool until the next billing cycle.
Overage behaviour¶
When the pool is exhausted, behaviour depends on the organization's additional-usage policy:
- Additional usage allowed — usage continues at published per-credit rates and is charged to the org/enterprise.
- Additional usage not allowed — usage is blocked until the next billing cycle.
If a user-level budget is set and a user exhausts it, that individual is cut off even if the org pool still has capacity. A $0 user-level budget means no Copilot access. There is no automatic fallback to cheaper models when a budget is hit — the request simply fails.
Budgets and cost control¶
Budgets can be set at four levels:
- Enterprise-level (all orgs, repos, cost centres)
- Organization-level
- Cost-centre-level
- User-level
Budgets are denominated in USD; consumption is reported in AI credits at the $0.01 conversion. Budgets can drive both alerts (approaching limit) and hard stops (block on limit).
Per-model pricing¶
Costs vary by model. Neowin's pre-announcement reporting cited example rates for a frontier model (e.g. ~$2.50 per million input tokens and ~$15 per million output tokens for a "GPT-5.4"-class model). The authoritative list lives in GitHub Docs under "Models and pricing for GitHub Copilot" — always consult that page for current numbers rather than the press reports.
What changed for individual plans¶
The org/enterprise pricing change happened alongside related individual-plan adjustments around the same period:
- Microsoft temporarily paused new sign-ups for Copilot Pro, Pro+, and Student plans in April 2026 to stabilize capacity for existing customers.
- Reduced usage limits were rolled out for individual plans.
- Claude Opus models were removed from the Pro plan tier.
The individual-plan ("Pro" / "Pro+" / "Student") pricing structure is not covered by the GitHub Docs page referenced here — that page applies only to Business and Enterprise SKUs. Pro plan token economics may still be in flux at time of writing.
Why this is happening¶
A few converging pressures:
- Agentic workloads consume orders of magnitude more tokens than chat. A coding agent doing a multi-file refactor with reasoning traces and tool calls can burn through what a year of chat would have used.
- Frontier model inference costs are high relative to the per-seat subscription price. Heavy agent users were costing vendors net-negative margin on a $19 seat.
- Industry alignment — Anthropic moved enterprise to token billing first; the rest of the market is following.
- Better cost attribution — usage-based billing lets enterprises see which teams / projects / users actually drive cost, which flat-rate hides.
Limitations / Caveats¶
- The Neowin source is a leak/preview from Ed Zitron's newsletter dated April 23, 2026 — confidence on the direction of change is high (corroborated by the GitHub Docs page), but the specific dollar figures in the Neowin piece (e.g. "$30 of pooled AI credits for Business at $19/month") are an early reading that doesn't exactly match the official docs ($19/month → 1,900 credits = $19 of credits, not $30 — the higher figure appears to conflate the standard allowance with the 3,000-credit promotional rate). Always treat the GitHub Docs page as authoritative over the Neowin numbers.
- No automatic fallback to cheaper models when a budget hits zero — admins must design budgets carefully to avoid abrupt productivity drops for engineers.
- Pooling means a single runaway user (e.g. an infinite agent loop) can drain a shared pool quickly if user-level budgets are not configured.
- Code completions are unlimited today — Microsoft has not committed to that being permanent.
- Per-model token pricing changes when new models ship; this page is a snapshot — verify against the live pricing doc.
- The promotional period ends September 1, 2026; budgets sized around the promo allowance will start failing in October without action.
Recommended actions¶
For org admins planning the transition:
- Audit current "premium request" usage per user — identify the top 10% of users; they are the credit-pool draw.
- Set organization-level alert budgets at 50% / 75% / 90% of expected monthly consumption.
- Set user-level hard budgets for the top consumers as a safety valve.
- Pick a default model policy — frontier models for agent work, cheaper models for chat — and document it for engineers.
- Plan a re-calibration for September 2026 when the promotional credits drop away.
Sources¶
- Usage-based billing for organizations and enterprises — GitHub Docs (authoritative reference, official Microsoft/GitHub docs)
- Report: GitHub Copilot is moving to token-based billing from June — Neowin, Apr 23 2026 (early tech-press reporting, useful for context on the why)
Changelog¶
- 2026-05-18 — Page created from the GitHub Docs "Usage-based billing for organizations and enterprises" reference page and the Neowin April 23 2026 report on Copilot's move to token-based billing. Corroboration between official docs and tech press raises confidence to high.