Azure AI Foundry no longer shows cached tokens for DeepSeek-V4.1-Flash (Global Standard)

I'm using DeepSeek-V4.1-Flash deployed via Microsoft Azure AI Foundry (Global Standard, model version 2026-09-10).

Until about two days ago, Azure Foundry's Monitor showed cached-token usage for this deployment, similar to my other model, GPT-6-Luna. Now, cached-token usage no longer appears, and the estimated cost has increased significantly (10x).

I'm using both DeepSeek and Luna through OpenCode, where cache hit rates are consistently above 95% for both models. Luna's Foundry Monitor still shows separate cost categories for cached tokens, while DeepSeek's does not.

This makes me suspect that caching is still working, but Azure Foundry may have stopped reporting or accounting for cached tokens correctly for DeepSeek.

Has there been a recent change or known issue with cached-token metering for DeepSeek-V4.1-Flash in Azure AI Foundry?
Is this a monitoring issue, a billing issue, or a change in the model's caching behavior?

Thanks!