GitHub Copilot AI Credits Explained: Token Pricing and the 27× Legacy Multiplier

Updated 25 July 2026: Most GitHub Copilot users moved to token-based GitHub AI Credits on 1 June 2026. The 27× Claude Opus multiplier and 9× Claude Sonnet multiplier now apply only to existing annual Copilot Pro and Pro+ subscribers who remained on legacy request-based billing until their current plan ends.

GitHub Copilot pricing is no longer best understood as a simple number of premium requests. Under GitHub's usage-based billing system, most paid users receive a monthly allowance of AI Credits, with chargeable activity calculated from model token prices.

A token is a small unit of text or code processed by an AI model. Longer prompts, larger files, extensive repository context and lengthy answers can therefore cost more than focused interactions, even when both appear to be a single request in the interface.

Who still pays the GitHub Copilot model multipliers?

The headline 27× figure belongs to GitHub's legacy request-based system. It is not the standard price calculation for a new Copilot subscriber or for an existing customer already moved to AI Credits.

Legacy multipliers apply only to existing annual Copilot Pro and Pro+ subscribers who chose to remain on request-based billing. For those subscribers, one model interaction can consume several premium requests according to the selected model's multiplier.

Legacy model	Request multiplier	Meaning
Claude Sonnet 4.6	9×	One interaction counts as nine premium requests
Claude Opus 4.6	27×	One interaction counts as 27 premium requests

Legacy multiplier table: These figures should not be applied to AI Credit allowances. GitHub documents them separately for qualifying annual plans that have not yet switched to the current billing model.

Current GitHub Copilot AI Credit allowances

For individual billing, one AI Credit equals US$0.01 of metered AI usage. The published individual totals combine fixed base credits with a flex allotment, so teams should use GitHub's current plan page when budgeting.

Plan	Included monthly AI Credits	Approximate value or pooling
Copilot Pro	1,500	US$15
Copilot Pro+	7,000	US$70
Copilot Max	20,000	US$200
Copilot Business	1,900 per assigned seat	Pooled at billing-entity level
Copilot Enterprise	3,900 per assigned seat	Pooled at billing-entity level

Existing Business and Enterprise customers receive temporary promotional allowances of 3,000 and 7,000 credits per assigned seat per month respectively, from 1 June to 1 September 2026. Organisations should avoid treating those promotional totals as their permanent operating baseline.

Paid plans continue to include unlimited code completions and next-edit suggestions. According to GitHub's individual billing documentation, those two features do not consume AI Credits.

What does one Copilot interaction cost?

There is no universal per-message price. GitHub calculates usage from the selected model's rates for input tokens, cached input, cache writes and output tokens. Cached input is previously processed context that the service can reuse more cheaply, while a cache write stores context for potential reuse.

Claude Sonnet 4.6 is listed at US$3 per million input tokens and US$15 per million output tokens. Claude Opus 4.6 is US$5 per million input tokens and US$25 per million output tokens. Both have separate, lower cached-input prices and separate cache-write rates.

A worked Copilot AI Credit example

An interaction using 10,000 uncached input tokens plus 2,000 output tokens costs about 6 AI Credits with Claude Sonnet 4.6:

• Input: 10,000 tokens at US$3 per million = US$0.03
• Output: 2,000 tokens at US$15 per million = US$0.03
• Total: US$0.06, equivalent to 6 AI Credits

The same token volumes cost about 10 AI Credits with Claude Opus 4.6:

• Input: 10,000 tokens at US$5 per million = US$0.05
• Output: 2,000 tokens at US$25 per million = US$0.05
• Total: US$0.10, equivalent to 10 AI Credits

These examples exclude cache writes and any additional calls made during an agent workflow. An apparently simple task can involve more than one model call when an agent searches files, uses tools, revises its work or continues autonomously.

Which Copilot features consume AI Credits?

Metered features include Copilot Chat, Copilot CLI, the cloud coding agent, Spaces, Spark and third-party coding agents. Code completions and next-edit suggestions remain outside AI Credit billing on paid plans.

Copilot code review needs particular attention. GitHub automatically selects an undisclosed model for the feature, and a review consumes both AI Credits and GitHub Actions minutes. Teams should monitor the combined cost rather than attempting to configure reviews around a named model.

This distinction matters when comparing AI coding tools with human developer costs. Subscription price alone does not capture metered model activity, automation infrastructure or the engineering time required to check generated work.

Monitoring usage and setting Copilot budgets

GitHub provides budgets at user, cost-centre, organisation and enterprise levels. A user-level budget caps that user's total consumption, while higher-level budgets generally control metered overage rather than removing the credits already included with a plan.

There is no automatic fallback to a cheaper model when a budget is exhausted and access is blocked. This is operationally important: a developer or workflow may stop rather than continue at reduced cost.

GitHub recommends matching model capability to the task, keeping context lean, preserving useful cache state and applying session limits. Its automatic model selection also provides a 10% discount on paid plans. These controls support a more deliberate form of agentic engineering rather than unstructured vibe coding.

A 30-day Copilot cost-control checklist

1. Days 1-3: Confirm whether each account uses AI Credits or the legacy annual request system.
2. Days 4-7: Record included credits, pooled organisation allowances and temporary promotional amounts separately.
3. Week 2: Identify usage from Chat, CLI, agents, Spaces, Spark and code review.
4. Week 2: Set user budgets for hard consumption limits and organisation budgets for overage control.
5. Week 3: Compare expensive-model usage with the actual complexity and value of each task.
6. Week 3: Reduce unnecessary repository context, repeated instructions and overlong agent sessions.
7. Week 4: Review AI Credit consumption alongside GitHub Actions minutes and developer review time.
8. Day 30: Set a sustainable baseline that excludes temporary promotional credits.

This is ultimately a compute-management problem, not just a licence-management problem. The same principle appears in the wider question of why AI compute is not automatically cheaper than people: cost depends on task design, utilisation, oversight and the value of the resulting work.

GitHub Copilot AI Credits FAQ

Does every Copilot user pay the 27× Claude Opus multiplier?

No. It applies only to eligible existing annual Pro and Pro+ subscribers who stayed on legacy request-based billing. Most users now pay through token-based AI Credits.

Do Copilot code completions consume AI Credits?

No. Paid plans include unlimited code completions and next-edit suggestions without deducting AI Credits.

Will Copilot switch to a cheaper model when my budget runs out?

No. GitHub states that a blocked user does not automatically fall back to a cheaper model.

Can I calculate the exact cost of code review from a named model?

No. GitHub automatically chooses the code review model and does not disclose which model is used. Code review also consumes GitHub Actions minutes as well as AI Credits.

What is the most useful number to monitor?

Track total AI Credits by feature and user, then review that alongside Actions minutes and engineering outcomes. A low token bill is not valuable if the work requires extensive correction, while higher usage may be justified for a difficult task that saves substantial developer time.