Finout Blog Archive

GitHub Copilot Is Now a Variable AI Cost. Here's How to Track It by Model

Written by Finout Writing Team | Oct 8, 2026, 2:30:56 PM

On October 19, GitHub retires six more Copilot models, including GPT-5.5, GPT-5.4 and GPT-5 mini. That follows six retirements on September 1, one on September 10 and four on October 2. In between, GitHub added eight models, from Gemini 3.8 Flash and GPT-6 Astra to Claude Opus 5.5 and GPT-6.1 Sol.

For the October 19 retirements, GitHub says the suggested replacements switch on automatically for Copilot Business and Enterprise, unless an admin has turned off default model enablement. Add auto model selection, which picks a model for each request, and your team's model mix can shift without anyone deciding it. Since Copilot charges for tokens at each model's own rate, your bill shifts with it.

That's why Finout now breaks down GitHub Copilot cost and usage by model. You can see which models your teams are using, what each one costs, and how the mix moves from day to day, next to the rest of your cloud and AI spend. Below, we cover why model churn is hard to track on a GitHub invoice and how the breakdown works.

Why the Model Now Sets the Price

We covered Copilot plans and per-model rates in detail in GitHub Copilot Pricing Plans, Hidden Costs, and 5 Ways to Control Them. The short version: since June 1, 2026, Copilot usage spends GitHub AI Credits, charged by tokens at each model's published rate. Code completions are still included. Seats come with a pool of credits, and once it runs out, usage is either billed at published rates or capped.

Then there's auto model selection. It isn't new: it has been generally available in VS Code since December 2025. Today, Copilot routes each task based on task complexity and system availability, across Chat, the CLI, the Copilot app and the cloud agent, and paid plans get a 10% discount on model costs for using it. What changed in June is that the price now tracks tokens at each model's rate. The model behind a request, and what that request costs, is often decided by Copilot rather than by anyone on your team.

Why It's Hard to See: The Model List Never Sits Still

If the model list were stable, you could learn your mix once and budget around it. It isn't, and it never has been. The difference now is that each swap can change your bill. Between September 1 and October 19, GitHub will have retired 17 Copilot models and added eight.

Date

Change

Models

Oct 19, 2026

Retiring

Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, Grok 4.5

Oct 2, 2026

Retired

Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code, Claude Opus 4.7

Sept 29, 2026

Added

GPT-6.1 Sol

Sept 28, 2026

Added

Claude Sonnet 5.5

Sept 22, 2026

Added

GPT-6 Sol, GPT-6 Luna

Sept 22, 2026

Added

Claude Opus 5.5

Sept 21, 2026

Added

Grok 4.7

Sept 10, 2026

Retired

MAI-Code-1-Flash

Sept 4, 2026

Added

GPT-6 Astra

Sept 3, 2026

Added

Gemini 3.8 Flash

Sept 1, 2026

Retired

Gemini 3.1 Pro, Claude Opus 4.5, Claude Opus 4.6, Claude Sonnet 4.5, Claude Sonnet 4.6, Raptor Mini

Every retirement comes with a suggested replacement, and the replacement can cost more or less than the model it replaces. Three of the October 19 swaps, at GitHub's published rates for standard context length:

Retiring model

Suggested replacement

Input, per 1M tokens

Output, per 1M tokens

GPT-5.4

GPT-5.6 Sol

$2.50 → $4.00

$15.00 → $20.00

GPT-5.5

GPT-5.6 Sol

$5.00 → $4.00

$30.00 → $20.00

GPT-5 mini

GPT-5.6 Luna

$0.25 → $0.20

$2.00 → $1.20

A team that leaned on GPT-5.4 pays 60% more per input token after the switch. A team on GPT-5.5 pays less. Same deprecation notice, opposite effects on the bill.

Three things make that shift hard to catch:

  1. It’s likely that nobody picked the model. With auto selection, there's no developer decision to review. The mix is an output, not an input.
  2. Spend gets reviewed monthly. If cost is checked when the invoice lands, a pricier model has been running for weeks by the time anyone notices.
  3. Copilot is often reviewed on its own. Copilot is billed by GitHub, not by your cloud or AI model providers. Unless your FinOps setup already pulls GitHub billing in next to those, Copilot gets looked at separately, and nobody sees a team's full AI spend across tools.

What FinOps Teams Need Now

Seat counts used to answer the Copilot cost question. Now you need to answer four more:

  • Which models are we paying for? Cost and AI-credit consumption per model, so you can see the mix and how it moves.
  • Who is driving it? Spend by organization, cost center and developer, so the cost lands with the team that created it.
  • What changed, and when? A daily view, not a monthly one, so a model swap shows up the week it happens.
  • How does it compare? Copilot next to your other AI and cloud spend, so each team's total AI cost sits in one place.

How Finout Shows Copilot Cost by Model

Finout's GitHub integration pulls billing data for Copilot, Actions, Codespaces, Packages, Git LFS and shared storage into MegaBill, alongside your cloud, Kubernetes, data platform and AI provider costs. Copilot Model is now one of its dimensions.

Cost and usage by model. Copilot usage carries a Copilot Model field, so you can group spend by model, SKU, organization and user. You see both net billed cost and list cost before discounts, and usage in AI credits next to dollars.

A ready-made GitHub dashboard. The predefined GitHub Costs dashboard shows total and projected GitHub spend, with Copilot views for SKU, AI credit consumption and seat usage by organization, plus top-spending users. There's nothing to configure once GitHub is connected.

One bill for AI spend. Copilot sits in the same MegaBill as your OpenAI, Anthropic, Cursor and cloud AI costs, so you can report on AI coding spend across tools instead of tool by tool.

Allocation to teams. Virtual Tags map GitHub organizations and users to the teams and business units you report on, so Copilot cost shows up in the same showback and chargeback as everything else.

Anomaly detection. Finout ships predefined anomalies for GitHub covering Actions, Copilot, SKU and organization, and you can add a custom anomaly grouped by Copilot Model. Alerts go to Slack, Microsoft Teams or email, so a jump in Copilot spend reaches you within days, not at month end.

Fresh data, with history. GitHub data refreshes daily, and each run re-reads the last seven days to pick up GitHub's retroactive adjustments. On first connection, Finout loads the current month plus up to the two previous full months.

Get Started

To connect to GitHub, you need a GitHub Enterprise Cloud account, a classic personal access token with the manage_billing:enterprise and manage_billing:copilot scopes, and your enterprise slug. Finout only reads billing and usage data; it never modifies your billing settings, Copilot seats, repositories or code. The full walkthrough is in the GitHub integration docs.

The next model will land in Copilot soon. With Finout, you'll see who is using it and what it costs within a day, not when the invoice arrives.

Book a demo to see Copilot cost by model in Finout.