I put $20 into DeepSeek and have been using it little by little.
Occasionally when I checked the charges, I was delighted to see that I hadn't used even $0.05.
I generated over 100 documents and tried various things for 2 weeks, and seeing that I still haven't spent even $2 total, I was happy about it when I received a notice that the charges are being doubled. It says it's from mid-July, and it seems like I could avoid it by using it only about 7 hours out of 24 hours. It overlaps with Korea's working hours, but looking closely, it seems to be based on Chinese working hours. It must be used a lot in China.
The following is the interpretation by Gemini.
The DeepSeek API service will introduce a peak-valley (peak hour differential) pricing plan starting from mid-July.
Peak hours will be charged at 2 times the regular rate.
This charge applies equally to all billing items (input tokens, output tokens, cache, and all applicable costs).
Peak hours (UTC)
1:00 AM ~ 4:00 AM
6:00 AM ~ 10:00 AM
Based on UTC+8 (China Standard Time)
9:00 AM ~ 12:00 PM
2:00 PM ~ 6:00 PM
Converted to Korea Time (KST, UTC+9)
Since Korea is UTC+9, it is 9 hours ahead of UTC.
That is, the peak pricing hours based on Korea time are
10:00 AM ~ 1:00 PM
3:00 PM ~ 7:00 PM
.
Meaning
If you are currently using the DeepSeek API,
10:00 AM~1:00 PM
3:00 PM~7:00 PM
API costs will be 2 times normal.
Conversely, if you schedule your work to avoid these times, you can reduce costs to about half.
In particular, for large batch jobs or agent (e.g., yb_os) nighttime automated research tasks, it is helpful for cost reduction to schedule them to run outside of peak hours.
