DeepSeek V4, which reached stable release on July 24, brought a change that went largely unnoticed by much of the market: double pricing during two daily peak windows — 9am-12pm and 2pm-6pm Beijing time (01:00-04:00 and 06:00-10:00 UTC). Outside those windows, pricing stays at V4 Pro's standard rates ($0.435 per million input tokens, $0.87 output) and V4 Flash's ($0.14 input, $0.28 output).
What Changed
According to reporting from TheRouter.ai and KuCoin, DeepSeek is now charging double the standard rate during Beijing's highest-demand hours, while keeping normal pricing outside those windows. The company notifies customers by email 24 hours before any rate change takes effect.
Why It Matters
For companies running AI workloads that don't need an instant response — batch processing, report generation, agent training, large-scale data analysis — the time of day a call is made now directly affects cost. Companies that ignore this detail pay up to double for no reason.
The Impact for Brazil
Here's the detail that matters for Brazilian companies: Beijing's peak hours fall, in Brasília time, between 10pm-1am and 3am-7am — outside Brazilian business hours. In practice, a Brazilian company running its batch AI workloads during business hours (8am-6pm Brasília time) is almost always already inside DeepSeek's cheaper pricing window, without doing anything differently. It's worth confirming the exact hours in the official documentation before relying on this operationally, since rates can change.
Entercast's Take
This is a concrete example of how AI cost savings don't come only from picking the right model — they also come from understanding a vendor's pricing architecture and designing the operation around it. Companies that treat AI cost as an engineering variable, not just a contract line item, find this kind of structural savings that cuts the cost of running AI at scale without giving up capability.