---
← Back to blog

DeepSeek V4-Flash just got more expensive

deepseekllmapi

The model my assistant runs on is deepseek-v4-flash. Yesterday DeepSeek announced API price increases for V4-Flash and V4-Pro, plus peak/off-peak billing, effective August 17, 00:00 Beijing time.

Flash used to be flat. Now there are two prices (official, per 1M tokens):

per 1M tokens Before After (off-peak) After (peak)
Input (cache miss) ¥1 ¥1.5 (+50%) ¥3 (+200%)
Output ¥2 ¥4.5 (+125%) ¥9 (+350%)
Input (cache hit) ¥0.02 ¥0.05 (+150%) ¥0.10 (+400%)

Peak hours are 9:00-12:00 and 14:00-18:00 Beijing time, everything else off-peak at half the peak rate. Which, where I live, is the entire working day.

So I pulled my usage export from the platform. Last 23 days: 29.6M fresh input tokens, 8.2M output, and 1.75 billion tokens read from cache. I run at a 98.3% cache hit rate, which was the whole point of Flash. The bill for that period: ¥81.23, which recomputes to the fen at official rates. And 62% of my cache reads land in peak hours, because I use it during working hours.

Monthly projection, same pattern:

per month Before After (off-peak) After (my mix) After (peak) OpenCode Go
Input (cache miss) ¥39 ¥58 ¥92 ¥116 included
Cache reads ¥46 ¥114 ¥185 ¥228 included
Output ¥21 ¥48 ¥79 ¥96 included
Total ¥106 ¥220 ¥356 ¥440 ¥71 flat

My bill goes from ¥106 to ¥356 a month. 3.4x, and it’s almost entirely the cache: 2.5x more expensive off-peak, 5x at peak, and I live in it.

Which makes the alternative hard to ignore: OpenCode Go still sells DeepSeek V4 Flash at the old rates, $10 a month flat (about ¥71), with $60 of usage included. My whole workload from that export burns about $15 of it. No peak, no off-peak. The price hike simply doesn’t reach it, at least until their rates catch up.

Details on the official pricing page. Still the cheapest frontier-class API around. The gap just closed a lot.