Last night, DeepSeek Issued a notice to its API users: Starting at 00:00 Beijing Time on August 23, the peak and off-peak pricing rules will be adjusted—during weekends (Saturdays and Sundays), the entire day will no longer distinguish between peak and off-peak periods, and all usage will be charged at the off-peak rate.
A single stone stirred up a thousand waves. This announcement sent shockwaves through the developer community and sparked quiet concerns among workers.
This is certainly good news for the broader developer community. Previously, DeepSeek V4-Flash and V4-Pro APIs have been switched to peak-off-peak pricing, with peak rates up to twice the off-peak rates. After this adjustment, developers no longer need to avoid peak hours on weekends—batch tasks can now be safely scheduled for weekends, resulting in noticeably lower bills.
Some users on X have directly said, "Finished a week's work in two days, so the bill was cut in half," but your actual bill will depend on your specific usage.
However, workers are beginning to worry: Will companies adjust work hours according to these peak and off-peak periods? Could it really happen that companies require work on weekends and offer weekday compensatory time off?
This may sound unbelievable, but it’s actually happening. The short video industry was among the first to implement this: recent reports showed that major platforms sequentially raised AI computing prices, causing the cost of tokens for short videos to nearly quadruple in a short period. Under this pressure, many AI short video production teams have begun scheduling their high-consumption rendering tasks for late at night, specifically to take advantage of cheaper computing rates during off-peak hours.
Soon, this trend spread to the programmer community. Recently, a programmer posted that their company implemented a new attendance policy to cut token costs, requiring employees to work one day on weekends and even adjusting lunch breaks to align with token peak and off-peak periods.
Someone remarked that each generation has its own script.
However, some have questioned whether this arrangement truly avoids additional communication costs—and whether it’s truly worthwhile.
There has also been speculation from outside about the reason for this price adjustment. Some say the peak price may correspond to DeepSeek The training time for the model, so the higher pricing is to reserve computing power for internal staff training. Over the weekend, DeepSeek Employees who aren't working have been switched to the valley price.
Some have questioned this claim, arguing that the model's training is highly automated and unrelated to whether one goes to work or not.
The new billing rules are now in effect. Did you start running batch tasks this Sunday as well?
Edited by Zhang Qian
This article is from the WeChat public account "Machine Heart" (ID: almosthuman2014), authored by Machine Heart.