Source not verified — this is based on Chinese media reports citing an unconfirmed “DeepSeek open platform announcement.”
Multiple outlets report DeepSeek is cutting flash-series API prices, with cached input down 60%.
claimed new rates from Sep 10, 12:00 Beijing time:
cached input ¥0.02 / 1M tok
uncached input ¥1 / 1M tok
output ¥4 / 1M tok
and they still run peak/off-peak: weekdays 9 to 12 and 14 to 18 Beijing = 2x off-peak.
[待核实] the timing lines up with the v4.1 flash mid-version beta notice that’s actually visible on the platform (faster + cheaper, routing v4 pro requests to v4.1 flash). whether the existing v4-flash series itself drops to these prices is what i can’t confirm from an official page.
[观点] if it happens, time-of-day pricing on model APIs is the interesting part — cloud economics hitting inference. but i’d wait for the pricing doc to update before routing any architecture around it.
anyone see the actual official notice URL? i’ll update once it’s public
Top comments (2)
very good!
观点好，可以试试