Copilot Studio billing overages due to wrong model charges

[Microsoft](https://www.linkedin.com/company/microsoft/) may have billed you for Claude Sonnet at Claude Opus rates - and you may not even know it. Yes, this is a real issue affecting #CopilotStudio. Microsoft has acknowledged two major issues related to cost reporting and licensing, both of which are fundamental to how customers manage and forecast their AI spend: 👉 Issue 6418847: Agents are being charged more starting in May despite no changes being made to them. 👉Issue 6472673: Claude Sonnet 4.6 is being incorrectly charged as a premium reasoning model. The first issue is relatively straightforward: Microsoft increased pricing, but the change was communicated poorly, resulting in unexpected consumption spikes for many customers. The second issue is more concerning. It was reported only two days ago and refers vaguely to "parts of May being affected," making it difficult to understand the actual impact. Personally, I've seen multiple agents that typically consumed 500–1,000 credits suddenly generate bills of 20,000–30,000 credits. Now I know why. These issues have had real consequences. Some customers were forced to pre-purchase additional credits under 12-month commitments. Others enabled uncontrolled pay-as-you-go billing, despite having limited visibility into whether charges were being calculated correctly. In some cases, organizations had to shut down entire environments or had automatically suspended all activity (including production) on tenants due to overages. I personally know of several such cases, and each of those can be summarized as pure chaos. This should also serve as a reminder that premium models can have dramatically different economics. Users experimenting with models such as Claude Opus 4.6, GPT-5.2 Reasoning, or future premium offerings should clearly understand that these models may generate costs that are 8–40× higher than standard-rate models such as GPT-5.3 Chat or Claude Sonnet 4.6. People naturally want to experiment with new models. The problem is that many organizations still lack sufficient visibility and education around AI cost management. An agent that normally costs $300 to run can suddenly generate a $10,000 bill. And yes, that's another real story. If you are an admin, start doing an audit on who uses premium models and prepare some ongoing communication. On a large scale, you may save hundreds of thousands of USD that way.