The short answer
GitHub's Copilot usage metrics API can now tell you which Copilot activity ran on Copilot-hosted models and which ran on your own provider key or a custom model. On November 3, 2026, GitHub added bring your own key (BYOK) fields to the enterprise and organization reports, at both user and aggregate level. For Copilot BYOK cost purposes, that means usage billed by your own model provider no longer has to be blended into a "Copilot cost per user" figure.
This matters because the two kinds of usage land on different invoices. Copilot-hosted usage is billed by GitHub, through seats and AI credits. A developer's own provider key is billed by that provider, and a local model on a laptop is billed by nobody. Before this release, all three could look identical in a report.
What GitHub changed
Per GitHub's changelog entry of November 3, 2026, the usage metrics API now reports two kinds of BYOK.
Local BYOK is configured by developers in their client. Requests go from Copilot CLI or the GitHub Copilot app straight to the developer's provider key or a local model, such as one served by Ollama, Foundry Local, LM Studio, or llama-server. The new fields include:
- used_local_byok: whether a user made at least one local BYOK request that day, in user-level reports.
- totals_by_local_byok: session, request, prompt, and token counts, split by byok_kind and by surface (Copilot CLI or the Copilot app).
- daily, weekly, and monthly active local BYOK user counts in aggregate reports.
Enterprise custom models are configured by admins in GitHub settings, and requests route through GitHub to your provider. They are flagged with a new is_custom_enterprise_model boolean on each model entry under totals_by_client. A model used both as a Copilot-hosted model and as a custom model appears as two entries, so the counts stay separate.
Both appear in 1-day and 28-day reports. Access is limited to enterprise owners, enterprise billing managers, organization owners, and custom roles with the View Copilot Metrics permission, and the Copilot usage metrics policy must be enabled.
Why Copilot BYOK cost was hard to see before
GitHub's own explanation is that BYOK usage used to blend into Copilot-hosted totals, because local and custom models often share names with Copilot models. The existing totals did not change with this release: BYOK was already counted, it is now just separable.
For a finance lead, the practical effect is on unit economics. If a team runs half its requests through its own provider key, a blended cost per request or per active user overstates what Copilot itself is billing and hides the spend that is going to a second vendor. The reverse also holds: a team that looks cheap per user may be cheap because its heavy usage sits on a provider invoice that nobody has attached to that team.
Read the numbers with these limits in mind
The changelog is explicit about where the new data is weaker, and these limits should shape how you use it.
- Local BYOK is a lower bound. The local fields rely on client telemetry, so they undercount when telemetry is disabled. Token counts can also undercount because some local providers do not report token usage.
- Enterprise custom models do not have that gap. GitHub observes those requests server-side, so the counts are real request and token counts.
- Local BYOK is not broken out by model name. Custom model names use existing name normalization, so unrecognized names appear as unknown.
- No backfill. Report days before the release do not include is_custom_enterprise_model, so you cannot rebuild history from it.
Two companion changes on November 5 help you judge how complete the data is. The API now flags users with missing client telemetry through client_telemetry_present and the daily missing_client_telemetry_user_count, and flags IDE or extension versions below the usage metrics minimums through usage_metrics_requirements and below_usage_metrics_minimum_user_count. A user can be counted in both, so do not add the two counts together, and null values mean the data is unavailable rather than zero.
That makes the telemetry-gap count a denominator-quality check. If a meaningful share of your users is missing telemetry, treat your local BYOK numbers as a floor, not a measurement.
A short checklist for finance and engineering
- Confirm the policy is on. The Copilot usage metrics policy has to be enabled before any of these fields appear.
- Pull the 28-day user report and look for used_local_byok. Count how many users send requests to their own keys or local models. Compare that against your BYOK policy.
- Check is_custom_enterprise_model for admin-configured models. These are the cleanest numbers, because they are observed server-side.
- Check the telemetry-gap count before trusting any local figure. High missing telemetry means the local BYOK picture is incomplete.
- Map provider invoices to the same teams. BYOK spend never appears on a GitHub bill. Your provider's invoice or console is where it lives, and it needs the same team attribution you use for Copilot. GitHub's cost center structure, covered in our guide to GitHub Copilot cost centers, can now also be queried by resource: a November 3 REST API addition returns the cost center a user, organization, repository, or enterprise team belongs to.
- Keep hosted and BYOK on separate lines in the forecast. One is a credit pool you can run out of, the other is a metered provider bill with its own rate card.
Where this fits in a wider spend picture
The new fields answer a Copilot-specific question: how much of the activity in my Copilot reports is actually Copilot's cost. They do not tell you what the BYOK usage cost, because the dollars sit with another provider. Getting to a complete figure means joining GitHub's report with provider billing data, and doing it again for every other assistant your teams use.
That is the general problem behind tracking AI code assistant spend across every vendor: each vendor exposes usage in its own shape, and a number is only comparable once it is normalized. Our walkthrough of the Copilot detailed usage report covers the per-user rows these new fields extend, and the guide to Copilot AI credits and premium request billing explains what the hosted side of the split actually bills. If you are building the forecast that sits on top, start with how to forecast AI coding spend.
Finance teams that want one cross-vendor view, with BYOK and hosted usage kept apart, can see how this fits the finance workflow. Engineering leaders who need to enforce a BYOK policy rather than just report on it will find the engineering view more relevant.
FAQ
Does BYOK usage show up on my GitHub invoice? The changelog describes local BYOK as requests sent directly to a developer's provider key or local model, and says usage billed by your own provider is separate from Copilot-hosted usage. For enterprise custom models, requests route through GitHub to your provider. Check your provider's billing for the charges themselves.
Can I see BYOK usage before November 3? Not through the new fields. GitHub says report days before the release do not include is_custom_enterprise_model, and BYOK was previously blended into hosted totals.
Are the local BYOK numbers exact? No. They rely on client telemetry, so they undercount when telemetry is disabled, and token counts can undercount where a local provider does not report them.
Who can read these reports? Enterprise owners, enterprise billing managers, organization owners, and anyone with a custom role that grants the View Copilot Metrics permission, provided the usage metrics policy is enabled.
![UI Practice - Duolingo [1]](https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F8uawvr6a0e07ilktpkk7.png)

