something seems not correct with ds v4 flash caching

Summary

Hi Command team,

Could you please help check whether there is an issue with DeepSeek V4 Flash caching or upstream routing on the Command servers?

My GOAT usage has increased unusually fast over the last couple of hours, and I hit the $35 weekly limit much faster than before, even though I have only been using deepseek/deepseek-v4-flash for past few hours and my workflow has not changed significantly.

One recent request from my usage dashboard shows:

• Input tokens: 357,723
• Output tokens: 120
• Charged: $0.0783
• Model: deepseek/deepseek-v4-flash

At the current off-peak DeepSeek V4 Flash pricing, 357,723 fresh input tokens × $0.22/M is approximately $0.0787, which is almost exactly what I was charged.

This seems to suggest that nearly the entire input was billed as fresh/cache-miss input rather than cache-read input.

This is unusual for my workload. When I previously used the DeepSeek API directly for the same type of long-running agent workflow, my prompt cache hit rate was typically around 95–98%. For example, on one day I had about 758M cached input tokens versus only 13.4M uncached input tokens.

Could you please check whether there has recently been any issue with:

• DeepSeek V4 Flash prompt caching
• upstream provider/server routing or cache affinity
• cache keys being invalidated between consecutive agent requests
• any changes related to the recent DeepSeek V4 Flash / pricing update

I’ve attached screenshots of both my Command usage and my previous DeepSeek API usage for comparison.

If possible, could you also check the trace IDs from my recent requests and confirm whether those requests were actually receiving prompt-cache hits?

Thanks!

Expected Behavior

n.a.

Actual Behavior

n.a.

Steps to reproduce the issue

n.a.

Command Code Version

claude code

Operating System

macOS

Terminal/IDE

No response

Shell

No response

Session file (optional)

No response

Fix prompt (optional)

No response

Additional context

No response

Bắt đầu bằng việc xem xét bằng chứng từ usage dashboard được đính kèm và các request trace IDs gần đây được đề cập trong báo cáo. So sánh input bị tính phí với hành vi cache-read dự kiến và xác minh xem các request DeepSeek V4 Flash có nhận được prompt-cache hits hay không; hoàn tất có nghĩa là xác nhận nguyên nhân do routing hoặc caching, hoặc loại trừ nguyên nhân đó bằng bằng chứng.