Bug Report: GPT-4.1 (0x) incorrectly charged as Premium Request (1x)

[NguyenVanHien12]

Hi everyone,

am reporting a potential billing/attribution bug in GitHub Copilot. Today (Jan 6, 2026), my usage of the GPT-4.1 model was incorrectly counted toward my Premium Request quota at a 1x multiplier, despite it being listed as an included (0x) model.

Steps to reproduce:

Select GPT-4.1 in Copilot Chat.

Send messages (only used Claude 3.5 Sonnet 3 times today).

Check the Usage Report.

Observed Result: My usage report (Account: NguyenVanHien12) shows 43 Premium Requests deducted for Jan 6th. Since I only used a premium model 3 times, approximately 40 requests from GPT-4.1 were wrongly attributed to the copilot_premium_request SKU.

Expected Result: GPT-4.1 usage should not consume Premium Request units (0x).

Evidence: I have downloaded my detailed CSV report and it confirms the discrepancy. (Attached a screenshot of the summary below).

summarizedUsageReport_1_e6f389051a014227bb9b61c5902a1afa.csv

[aguiarcode]

Hi there! Certain Copilot Chat interactions can still be counted as Premium Requests depending on:

• the feature surface used (e.g., Chat vs. inline suggestions),

• whether the request involved tool/function calls,

• or whether the request was routed through a premium Copilot backend path.

In other words, the model name alone does not always determine billing attribution.

Usage reports aggregate by SKU, not by visible model choice:
The copilot_premium_request SKU reflects how the request was processed internally, not necessarily the model you explicitly selected in the UI. It’s possible for requests that appear to use GPT-4.1 to still be billed under a premium SKU if they traverse a premium execution path.

I’d recommend opening a GitHub Support ticket and attaching the CSV directly. Support can verify whether this is expected attribution behavior or a genuine billing bug and adjust usage if needed.

[NguyenVanHien12]

Thank you for the response, but I believe this explanation does not fit my case.

I understand that 'premium paths' exist, but having 40 out of 43 requests charged as premium while using GPT-4.1 is not a minor attribution issue—it's a complete failure of the 0x model promise.

User Expectation: If the UI explicitly labels GPT-4.1 as 0x (Included), users should be able to rely on that. If 95% of my '0x' chats are silently routed through 'premium paths' and consume my quota, the 0x label is misleading.

Simple Usage: My interactions were standard chat queries. I did not use complex extensions or specialized tools that would justify such heavy premium usage.

Transparency: If a specific chat is going to be billed as premium despite selecting a 0x model, there should be a real-time warning in the UI.

I have already opened a support ticket with the CSV attached as suggested. I am posting here to see if other users are experiencing this 'silent' quota consumption, as it makes the Premium Request limit impossible to manage

[aguiarcode]

That’s a fair pushback, and I agree with the concern you’re raising.
At this point, you’re doing the right thing by involving Support with concrete CSV evidence. From a community perspective, this is the kind of feedback that actually helps clarify where expectations and behavior are misaligned.

[NguyenVanHien12]

Thank you for the explanation. However, I believe this is a genuine billing bug rather than expected behavior for the following reasons: 1. *Massive Discrepancy:* I used Claude 4.5 Sonnet (1x) only *3 times*. The report shows *43 premium requests*. This means 40 requests using GPT-4.1 were charged as premium. 2. *Standard Usage:* I was performing standard Chat interactions without complex tool/function calls or Extensions that would justify a "premium execution path" for 40 consecutive messages. 3. *Consistency:* I have used GPT-4.1 in the exact same way previously without any premium request deductions. This issue only occurred on Jan 6th. I have already attached the CSV report to the original ticket. Please escalate this to a billing specialist to review the internal logs for my account (*NguyenVanHien12*) for Jan 6th. If GPT-4.1 is marketed as 0x in the UI, charging it as 1x without clear warning is a misleading user experience. Please verify and credit back the 40 requests. Vào Thứ 3, 6 thg 1, 2026 vào lúc 16:50 João ***@***.***> đã viết:

…

Hi there! Certain Copilot Chat interactions can still be counted as Premium Requests depending on: - the feature surface used (e.g., Chat vs. inline suggestions), - whether the request involved tool/function calls, - or whether the request was routed through a premium Copilot backend path. In other words, the model name alone does not always determine billing attribution. Usage reports aggregate by SKU, not by visible model choice: The copilot_premium_request SKU reflects how the request was processed internally, not necessarily the model you explicitly selected in the UI. It’s possible for requests that appear to use GPT-4.1 to still be billed under a premium SKU if they traverse a premium execution path. I’d recommend opening a GitHub Support ticket and attaching the CSV directly. Support can verify whether this is expected attribution behavior or a genuine billing bug and adjust usage if needed. — Reply to this email directly, view it on GitHub <#183596 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BBH6ILSSBOQF5YTNSCIXA5L4FOAM7AVCNFSM6AAAAACQZ3CNXCVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTKNBSGI4DAOI> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[NguyenVanHien12]

"Hi Support, I raised this issue on GitHub Discussions, and the community/staff moderator there agreed that my concern is a 'fair pushback' regarding the misalignment between the UI (GPT-4.1 labeled as 0x) and the actual billing behavior (43 requests deducted). You can see the discussion here: [Dán link bài Discussion của bạn vào đây] Given that the community staff has acknowledged this discrepancy, I request that you prioritize my ticket and credit back the 40 requests that were incorrectly deducted. The CSV evidence clearly shows this was not a case of a few 'premium paths', but a systematic misbilling of the GPT-4.1 model for my account. Looking forward to your resolution." Vào Thứ 3, 6 thg 1, 2026 vào lúc 21:39 João ***@***.***> đã viết:

…

That’s a fair pushback, and I agree with the concern you’re raising. At this point, you’re doing the right thing by involving Support with concrete CSV evidence. From a community perspective, this is the kind of feedback that actually helps clarify where expectations and behavior are misaligned. — Reply to this email directly, view it on GitHub <#183596 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BBH6ILXEGUMEWEMY7WEDHGT4FPCJ3AVCNFSM6AAAAACQZ3CNXCVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTKNBSGUZDCMA> . You are receiving this because you authored the thread.Message ID: ***@***.***>